You are on page 1of 476

Cloudera Administration

Important Notice
(c) 2010-2016 Cloudera, Inc. All rights reserved.
Cloudera, the Cloudera logo, Cloudera Impala, and any other product or service
names or slogans contained in this document are trademarks of Cloudera and its
suppliers or licensors, and may not be copied, imitated or used, in whole or in part,
without the prior written permission of Cloudera or the applicable trademark holder.

Hadoop and the Hadoop elephant logo are trademarks of the Apache Software
Foundation. All other trademarks, registered trademarks, product names and
company names or logos mentioned in this document are the property of their
respective owners. Reference to any products, services, processes or other
information, by trade name, trademark, manufacturer, supplier or otherwise does
not constitute or imply endorsement, sponsorship or recommendation thereof by
us.
Complying with all applicable copyright laws is the responsibility of the user. Without
limiting the rights under copyright, no part of this document may be reproduced,
stored in or introduced into a retrieval system, or transmitted in any form or by any
means (electronic, mechanical, photocopying, recording, or otherwise), or for any
purpose, without the express written permission of Cloudera.
Cloudera may have patents, patent applications, trademarks, copyrights, or other
intellectual property rights covering subject matter in this document. Except as
expressly provided in any written license agreement from Cloudera, the furnishing
of this document does not give you any license to these patents, trademarks
copyrights, or other intellectual property. For information about patents covering
Cloudera products, see http://tiny.cloudera.com/patents.
The information in this document is subject to change without notice. Cloudera
shall not be liable for any damages resulting from technical errors or omissions
which may be present in this document, or from use of this document.
Cloudera, Inc.
1001 Page Mill Road, Bldg 3
Palo Alto, CA 94304
info@cloudera.com
US: 1-888-789-1488
Intl: 1-650-362-0488
www.cloudera.com
Release Information
Version: 5.5.x
Date: January 14, 2016

Table of Contents
About Cloudera Administration................................................................................7
Managing CDH and Managed Services....................................................................8
Managing CDH and Managed Services Using Cloudera Manager..........................................................8
Configuration Overview............................................................................................................................................8
Managing Clusters..................................................................................................................................................32
Managing Services..................................................................................................................................................35
Managing Roles......................................................................................................................................................44
Managing Hosts......................................................................................................................................................49
Maintenance Mode.................................................................................................................................................62

Managing CDH from the Command Line................................................................................................64


Starting CDH Services............................................................................................................................................65
Stopping CDH Services Using the Command Line..............................................................................................70
Migrating Data between a CDH 4 and CDH 5 Cluster.........................................................................................72

Managing Individual Services...................................................................................................................77


Managing Flume.....................................................................................................................................................77
Managing HBase....................................................................................................................................................79
Managing HDFS....................................................................................................................................................124
Managing Hive......................................................................................................................................................157
Managing Hue.......................................................................................................................................................166
Managing Impala..................................................................................................................................................189
Managing Key-Value Store Indexer....................................................................................................................201
Managing MapReduce and YARN.......................................................................................................................203
Managing Oozie....................................................................................................................................................210
Managing Solr.......................................................................................................................................................218
Managing Spark....................................................................................................................................................220
Managing the Sqoop 1 Client..............................................................................................................................242
Managing Sqoop 2................................................................................................................................................243
Managing ZooKeeper...........................................................................................................................................243
Configuring Services to Use the GPL Extras Parcel..........................................................................................244

Resource Management.........................................................................................246
Schedulers................................................................................................................................................246
Cloudera Manager Resource Management Features.........................................................................246
Linux Control Groups (cgroups)..............................................................................................................248
Resource Management with Control Groups....................................................................................................250

Configuring Resource Parameters.....................................................................................................................250

Static Service Pools.................................................................................................................................252


Dynamic Resource Pools........................................................................................................................253
Managing Dynamic Resource Pools...................................................................................................................253
YARN Pool Status and Configuration Options..................................................................................................256
Assigning Applications and Queries to Resource Pools...................................................................................257
Configuration Sets................................................................................................................................................259
Scheduling Rules..................................................................................................................................................260

Managing Impala Admission Control....................................................................................................261


Managing the Impala Llama ApplicationMaster..................................................................................263
Enabling Integrated Resource Management Using Cloudera Manager.........................................................264
Disabling Integrated Resource Management Using Cloudera Manager........................................................264
Configuring Llama Using Cloudera Manager.....................................................................................................265

Impala Resource Management..............................................................................................................265


Admission Control and Query Queuing..............................................................................................................265
Integrated Resource Management with YARN.................................................................................................273

Performance Management...................................................................................277
Optimizing Performance in CDH............................................................................................................277
Choosing a Data Compression Format.................................................................................................280
Tuning the Solr Server.............................................................................................................................281
Tuning to Complete During Setup.......................................................................................................................281
General Tuning......................................................................................................................................................282
Other Resources...................................................................................................................................................286

Tuning Spark Applications......................................................................................................................286


Optimizing Spark Performance...........................................................................................................................286

High Availability.....................................................................................................288
HDFS High Availability............................................................................................................................288
Introduction to HDFS High Availability..............................................................................................................288
Configuring Hardware for HDFS HA...................................................................................................................290
Enabling HDFS HA................................................................................................................................................291
Disabling and Redeploying HDFS HA.................................................................................................................304
Configuring Other CDH Components to Use HDFS HA.....................................................................................308
Administering an HDFS High Availability Cluster.............................................................................................310
Changing a Nameservice Name for Highly Available HDFS Using Cloudera Manager.................................314

MapReduce (MRv1) and YARN (MRv2) High Availability.....................................................................315


YARN (MRv2) ResourceManager High Availability...........................................................................................315
Work Preserving Recovery for YARN Components...........................................................................................322
MapReduce (MRv1) JobTracker High Availability..............................................................................................324

Cloudera Navigator Key Trustee Server High Availability...................................................................337


Configuring Key Trustee Server High Availability Using Cloudera Manager..................................................337

Configuring Key Trustee Server High Availability Using the Command Line.................................................338
Recovering a Key Trustee Server........................................................................................................................339

High Availability for Other CDH Components.......................................................................................339


HBase High Availability........................................................................................................................................339
Hue High Availability ...........................................................................................................................................344
Hive Metastore High Availability........................................................................................................................347
Llama High Availability........................................................................................................................................349
Configuring Oozie for High Availability...............................................................................................................351
Search High Availability.......................................................................................................................................352

Configuring Cloudera Manager for High Availability With a Load Balancer......................................353


Introduction to Cloudera Manager Deployment Architecture.........................................................................354
Prerequisites for Setting up Cloudera Manager High Availability...................................................................355
High-Level Steps to Configure Cloudera Manager High Availability ..............................................................355
Database High Availability Configuration..........................................................................................................381
TLS and Kerberos Configuration for Cloudera Manager High Availability......................................................382

Backup and Disaster Recovery.............................................................................384


Port Requirements for Backup and Disaster Recovery.......................................................................384
Data Replication......................................................................................................................................385
Designating a Replication Source.......................................................................................................................388
HDFS Replication..................................................................................................................................................389
Hive Replication....................................................................................................................................................395
Impala Metadata Replication..............................................................................................................................402
Using Snapshots with Replication......................................................................................................................402
Enabling Replication Between Clusters in Different Kerberos Realms..........................................................403
Replication of Encrypted Data............................................................................................................................404
HBase Replication................................................................................................................................................405

Snapshots................................................................................................................................................410
Cloudera Manager Snapshot Policies.................................................................................................................410
Managing HBase Snapshots...............................................................................................................................414
Managing HDFS Snapshots................................................................................................................................424

Cloudera Manager Administration......................................................................429


Starting, Stopping, and Restarting the Cloudera Manager Server.....................................................429
Configuring Cloudera Manager Server Ports........................................................................................429
Moving the Cloudera Manager Server to a New Host.........................................................................429
Managing the Cloudera Manager Server Log.......................................................................................430
Viewing the Log....................................................................................................................................................430
Setting the Cloudera Manager Server Log Location.........................................................................................430

Cloudera Manager Agents......................................................................................................................431


Starting, Stopping, and Restarting Cloudera Manager Agents........................................................................431
Configuring Cloudera Manager Agents..............................................................................................................433

Managing Cloudera Manager Agent Logs..........................................................................................................436

Changing Hostnames..............................................................................................................................437
Configuring Network Settings................................................................................................................439
Alerts.........................................................................................................................................................439
Managing Alerts...................................................................................................................................................440

Managing Licenses..................................................................................................................................447
Sending Usage and Diagnostic Data to Cloudera.................................................................................452
Configuring a Proxy Server..................................................................................................................................452
Managing Anonymous Usage Data Collection..................................................................................................452
Managing Hue Analytics Data Collection...........................................................................................................452
Diagnostic Data Collection...................................................................................................................................453

Exporting and Importing Cloudera Manager Configuration................................................................455


Backing up Cloudera Manager...............................................................................................................455
Backing up Databases.........................................................................................................................................457

Other Cloudera Manager Tasks and Settings.......................................................................................457


Settings..................................................................................................................................................................457
Alerts......................................................................................................................................................................458
Users......................................................................................................................................................................458
Kerberos................................................................................................................................................................458
License...................................................................................................................................................................458
User Interface Language......................................................................................................................................458
Peers......................................................................................................................................................................458

Cloudera Management Service..............................................................................................................458

Cloudera Navigator Data Management Component Administration..............464


Cloudera Navigator Audit Server............................................................................................................464
Publishing Audit Events......................................................................................................................................467

Cloudera Navigator Metadata Server....................................................................................................469

About Cloudera Administration

About Cloudera Administration


This guide describes how to configure and administer a Cloudera deployment. Administrators manage resources,
availability, and backup and recovery configurations. In addition, this guide shows how to implement high
availability, and discusses integration.

Cloudera Administration | 7

Managing CDH and Managed Services

Managing CDH and Managed Services


If you use Cloudera Manager to manage your cluster, configuring and managing your cluster, as well as individual
services and hosts, uses a different paradigm than if you use CDH without Cloudera Manager. For this reason,
many of these configuration tasks offer two different subtasks, one each for clusters managed by Cloudera
Manager and one for clusters which do not use Cloudera Manager. Often, the tasks are not interchangeable.
For instance, if you use Cloudera Manager you cannot use standard Hadoop command-line utilities to start and
stop services. Instead, you use Cloudera Manager to perform these tasks.

Managing CDH and Managed Services Using Cloudera Manager


You manage CDH and managed services using the Cloudera Manager Admin Console and Cloudera Manager
API.
The following sections focus on the Cloudera Manager Admin Console.

Configuration Overview
When Cloudera Manager configures a service, it allocates roles that are required for that service to the hosts in
your cluster. The role determines which service daemons run on a host.
For example, for an HDFS service instance, Cloudera Manager configures:

One host to run the NameNode role.


One host to run as the secondary NameNode role.
One host to run the Balancer role.
Remaining hosts as to run DataNode roles.

A role group is a set of configuration properties for a role type, as well as a list of role instances associated with
that group. Cloudera Manager automatically creates a default role group named Role Type Default Group for
each role type.
When you run the installation or upgrade wizard, Cloudera Manager configures the default role groups it adds,
and adds any other required role groups for a given role type. For example, a DataNode role on the same host
as the NameNode might require a different configuration than DataNode roles running on other hosts. Cloudera
Manager creates a separate role group for the DataNode role running on the NameNode host and uses the
default configuration for DataNode roles running on other hosts.
Cloudera Manager wizards autoconfigure role group properties based on the resources available on the hosts.
For properties that are not dependent on host resources, Cloudera Manager default values typically align with
CDH default values for that configuration. Cloudera Manager deviates when the CDH default is not a recommended
configuration or when the default values are illegal.
Cloudera Manager Configuration Layout
After running the Installation wizard, use Cloudera Manager to reconfigure the existing services and add and
configure additional hosts and services.
Cloudera Manager configuration screens offer two layout options: new (the default) and classic. You can switch
between layouts using the Switch to XXX layout link at the top right of the page. Keep the following in mind
when you select a layout:
If you switch to the classic layout, Cloudera Manager preserves that setting when you upgrade to a new
version.
Selections made in one layout are not preserved when you switch.

8 | Cloudera Administration

Managing CDH and Managed Services


Certain features, including controls for configuring Navigator audit events and HDFS log redaction, are
supported only in the new layout.
New layout pages contain controls that allow you filter configuration properties based on configuration status,
category, and group. For example, to display the JournalNode maximum log size property (JournalNode Max Log
Size), click the CATEGORY > JournalNode and GROUP > Logs filters:

When a configuration property has been set to a value different from the default, a reset to default value icon
displays.
Classic layout pages are organized by role group and categories within the role group. For example, to display
the JournalNode maximum log size property (JournalNode Max Log Size), select JournalNode Default Group >
Logs.

Cloudera Administration | 9

Managing CDH and Managed Services

When a configuration property has been set to a value different from the default, a Reset to the default value
link displays.
There is no mechanism for resetting to an autoconfigured value. However, you can use the configuration history
and rollback feature to revert any configuration changes.

Modifying Configuration Properties Using Cloudera Manager


Note:
This topic discusses how to configure properties using the Cloudera Manager "new layout." The older
layout, called the "classic layout" is still available. For instructions on using the classic layout, see
Modifying Configuration Properties (Classic Layout) on page 15.
To switch between the layouts, click either the Switch to the new layout or Switch to the classic layout
links in the upper-right portion of all configuration pages.

Required Role:
When a service is added to Cloudera Manager, either through the installation or upgrade wizard or with the Add
Services workflow, Cloudera Manager automatically sets the configuration properties, based on the needs of
the service and characteristics of the cluster in which it will run. These configuration properties include both
service-wide configuration properties, as well as specific properties for each role type associated with the service,
managed through role groups. A role group is a set of configuration properties for a role type, as well as a list
of role instances associated with that group. Cloudera Manager automatically creates a default role group named
Role Type Default Group for each role type. See Role Groups on page 47.

10 | Cloudera Administration

Managing CDH and Managed Services


Changing the Configuration of a Service or Role Instance
1. Go to the service status page. (Cluster > service name)
2. Click the Configuration tab.
3. Locate the property you want to edit. You can type all or part of the property name in the search box, or use
the filters on the left side of the screen:
The Status section limits the displayed properties by their status. Possible statuses include:

Error
Warning
Edited
Non-default
Has Overrides

The Scope section of the left hand panel organizes the configuration properties by role types; first those
that are Service-Wide, followed by various role types within the service. When you select one of these
roles, a set of properties whose values are managed by the default role group for the role display. Any
additional role groups that apply to the property also appear in this panel and you can modify values for
each role group just as you can the default role group.
The Category section of the left hand panel allows you to limit the displayed properties by category.
4. Edit the property value.
To facilitate entering some types of values, you can specify not only the value, but also the units that
apply to the value. for example, to enter a setting that specifies bytes per second, you can choose to enter
the value in bytes (B), KiBs, MiBs, or GiBsselected from a drop-down menu that appears when you edit
the value.
If the property allows a list of values, click the icon to the right of the edit field to add an additional
field. An example of this is the HDFS DataNode Data Directory property, which can have a comma-delimited
list of directories as its value. To remove an item from such a list, click the icon to the right of the field
you want to remove.
Many configuration properties have different values that are configured by multiple role groups. (See Role
Groups on page 47).
To edit configuration values for multiple role groups:
1. Navigate to the property, For example, the configuration panel for the Heap Dump Directory property
displays the DataNode Default Group (a role group), and a link that says ... and 6 others.

2. Click the ... and 6 others link to display all of the role groups:

3. Click the Show fewer link to collapse the list of role groups.
If you edit the single value for this property, Cloudera Manager applies the value to all role groups. To edit
the values for one or more of these role groups individually, click Edit Individual Values. Individual fields
display where you can edit the values for each role group. For example:
Cloudera Administration | 11

Managing CDH and Managed Services

5. Click Save Changes to commit the changes. You can add a note that is included with the change in the
Configuration History. This changes the setting for the role group, and applies to all role instances associated
with that role group. Depending on the change you made, you may need to restart the service or roles
associated with the configuration you just changed. Or, you may need to redeploy your client configuration
for the service. You should see a message to that effect at the top of the Configuration page, and services
will display an outdated configuration (Restart Needed), (Refresh Needed), or outdated client configuration
indicator. Click the indicator to display the Stale Configurations on page 28 page.

Searching for Properties


You can use the Search box to search for properties by name or label. The search also returns properties whose
description matches your search term.

Validation of Configuration Properties


Cloudera Manager validates the values you specify for configuration properties. If you specify a value that is
outside the recommended range of values or is invalid, Cloudera Manager displays a warning at the top of the
Configuration tab and in the text box after you click Save Changes. The warning is yellow if the value is outside
the recommended range of values and red if the value is invalid.

Overriding Configuration Properties


For role types that allow multiple instances, each role instance inherits its configuration properties from its
associated role group. While role groups provide a convenient way to provide alternate configuration properties
for selected groups of role instances, there may be situations where you want to make a one-off configuration
changefor example when a host has malfunctioned and you want to temporarily reconfigure it. In this case,
you can override configuration properties for a specific role instance:
1.
2.
3.
4.
5.
6.

Go to the Status page for the service whose role you want to change.
Click the Instances tab.
Click the role instance you want to change.
Click the Configuration tab.
Change the configuration values as appropriate.
Save your changes.

You will most likely need to restart your service or role to have your configuration changes take effect. See Stale
Configuration Actions on page 29.

Viewing and Editing Overridden Configuration Properties


To see a list of all role instances that have an override value for a particular configuration setting, navigate to
the Status page for the service and select Status > Has overrides. A list of configuration properties where values
have been overridden displays. The panel for each configuration property displays the values for each role group
12 | Cloudera Administration

Managing CDH and Managed Services


or instance. You can edit the value of this property for this instance, or, you can click the icon next to an instance
name to remove the overridden value.

Resetting Configuration Properties to the Default Value


To reset a property back to its default value, click the

icon. The default value is inserted and the icon turns

into an Undo icon ( ). Explicitly setting a configuration to the same value as its default (inherited value) has
the same effect as using the
icon.

There is no mechanism for resetting to an autoconfigured value. However, you can use the configuration history
and rollback feature to revert any configuration changes.

Viewing and Editing Host Overrides


You can override the properties of individual hosts in your cluster.
1.
2.
3.
4.

Click the Hosts tab.


Click the Configuration tab.
Use the Filters or Search box to locate the property that you want to override.
Click the Manage Host Overrides link.

The Manage Overrides dialog box displays.


5. Select one or more hosts to override this property.
6. Click Update.
A new entry area displays where you can enter the override values. In the example below, servers
ed9-e.ent.cloudera.com and ed9-r.cloudera.com were selected for overrides. Note that the first set
of fields displays the value set for all hosts and the two sets of fields that follow allow you to edit the override
values for each specified host.

Cloudera Administration | 13

Managing CDH and Managed Services

To remove the override, click the

icon next to the host name.

To apply the same value to all hosts, click Edit Identical Values. Click Edit Individual Values to apply different
values to selected hosts.
7. If the property indicates Requires Agent Restart, restart the agent on the affected hosts.
Restarting Services and Instances after Configuration Changes
If you change the configuration properties after you start a service or instance, you may need to restart the
service or instance to have the configuration properties become active. If you change configuration properties
at the service level that affect a particular role only (such as all DataNodes but not the NameNodes), you can
restart only that role; you do not need to restart the entire service. If you changed the configuration for a particular
role instance (such as one of four DataNodes), you may need to restart only that instance.
1. Follow the instructions in Restarting a Service on page 41 or Starting, Stopping, and Restarting Role Instances
on page 46.
2. If you see a Finished status, the service or role instances have restarted.
3. Navigate to the Home > Status tab. The service should show a Status of Started for all instances and a health
status of Good.
For further information, see Stale Configurations on page 28.
Suppressing Configuration and Parameter Validation Warnings
You can suppress the warnings that Cloudera Manager issues when a configuration value is outside the
recommended range or is invalid. If a warning does not apply to your deployment, you might want to suppress
it. Suppressed validation warnings are still retained by Cloudera Manager, and you can unsuppress the warnings
at any time. You can suppress each warning when you view it, or you can configure suppression for a specific
validation before warnings occur.
Suppressing a Configuration Validation in Cloudera Manager
1. Click the Suppress... link to suppress the warning.
A dialog box opens where you can enter a comment about the suppression.
2. Click Confirm.
You can also suppress warnings from the All Configuration Issues screen:
1. Browse to the Home screen.
2. Click Configurations > Configuration Issues.
3. Locate the validation message in the list and click the Suppress... link.
A dialog box opens where you can enter a comment about the suppression.
4. Click Confirm.
The suppressed validation warning is now hidden.
14 | Cloudera Administration

Managing CDH and Managed Services


Managing Suppressed Validations
On pages where you have suppressed validations, you see a link that says Show # Suppressed Warning(s). On
this screen, you can:
Click the Show # Suppressed Warning(s) link to show the warnings.
Each suppressed warning displays an icon:

Click the Unsuppress... link to unsuppress the configuration validation.


Click the Hide Suppressed Warnings link to re-hide the suppressed warnings.
Suppressing Configuration Validations Before They Trigger Warnings
1. Navigate to the service or host with the configuration validation warnings you want to suppress.
2. Click Configuration.
3. In the filters on the left, select Category > Suppressions.
A list of suppression properties displays. The names of the properties begin with Suppress Parameter
Validation or Suppress Configuration Validator. You can also use the Search function to limit the number of
properties that display.
4. Select a suppression property to suppress the validation warning.
5. Click Save Changes to commit the changes.
Viewing a List of All Suppressed Validations
Do one of the following:
From the Home page or the Status page of a cluster, select Configuration > Suppressed Health and
Configuration Issues.
From the Status page of a service, select Configuration > Category > Suppressions and select Status >
Non-default.
From the Host tab, select Configuration > Category > Suppressions and select Status > Non-default.

Modifying Configuration Properties (Classic Layout)


Required Role:
Note: As of Cloudera Manager version 5.2, a new layout of the pages where you configure Cloudera
Manager system properties was introduced. In Cloudera Manager version 5.4, this new layout displays
by default. This topic discusses how to configure properties using the older layout, called the "Classic
Layout". For instructions on using the new layout, see Modifying Configuration Properties Using
Cloudera Manager on page 10.
To switch between the layouts, click either the Switch to the new layout or Switch to the classic layout
links in the upper-right portion of all configuration pages.
When a service is added to Cloudera Manager, either through the installation or upgrade wizard or with the Add
Services workflow, Cloudera Manager automatically sets the configuration properties, based on the needs of
the service and characteristics of the cluster in which it will run. These configuration properties include both
service-wide configuration properties, as well as specific properties for each role type associated with the service,
managed through role groups. A role group is a set of configuration properties for a role type, as well as a list
of role instances associated with that group. Cloudera Manager automatically creates a default role group named
Role Type Default Group for each role type. See Role Groups on page 47.
Changing the Configuration of a Service or Role Instance (Classic Layout)
1. Go to the service status page.
2. Click the Configuration tab.
Cloudera Administration | 15

Managing CDH and Managed Services


3. Under the appropriate role group, select the category for the properties you want to change.
4. To search for a text string (such as "snippet"), in a property, value, or description, enter the text string in the
Search box at the top of the category list.
5. Moving the cursor over the value cell highlights the cell; click anywhere in the highlighted area to enable
editing of the value. Then type the new value in the field provided (or check or uncheck the box, as appropriate).
To facilitate entering some types of values, you can specify not only the value, but also the units that
apply to the value. for example, to enter a setting that specifies bytes per second, you can choose to enter
the value in bytes (B), KiBs, MiBs, or GiBsselected from a drop-down menu that appears when you edit
the value.
If the property allows a list of values, click the icon to the right of the edit field to add an additional
field. An example of this is the HDFS DataNode Data Directory property, which can have a comma-delimited
list of directories as its value. To remove an item from such a list, click the icon to the right of the field
you want to remove.
6. Click Save Changes to commit the changes. You can add a note that will be included with the change in the
Configuration History. This will change the setting for the role group, and will apply to all role instances
associated with that role group. Depending on the change you made, you may need to restart the service or
roles associated with the configuration you just changed. Or, you may need to redeploy your client configuration
for the service. You should see a message to that effect at the top of the Configuration page, and services
will display an outdated configuration (Restart Needed), (Refresh Needed), or outdated client configuration
indicator. Click the indicator to display the Stale Configurations on page 28 page.
Validation of Configuration Properties
Cloudera Manager validates the values you specify for configuration properties. If you specify a value that is
outside the recommended range of values or is invalid, Cloudera Manager displays a warning at the top of the
Configuration tab and in the text box after you click Save Changes. The warning is yellow if the value is outside
the recommended range of values and red if the value is invalid.
Overriding Configuration Properties
For role types that allow multiple instances, each role instance inherits its configuration properties from its
associated role group. While role groups provide a convenient way to provide alternate configuration properties
for selected groups of role instances, there may be situations where you want to make a one-off configuration
changefor example when a host has malfunctioned and you want to temporarily reconfigure it. In this case,
you can override configuration properties for a specific role instance:
1.
2.
3.
4.
5.
6.

Go to the Status page for the service whose role you want to change.
Click the Instances tab.
Click the role instance you want to change.
Click the Configuration tab.
Change the configuration values as appropriate.
Save your changes.

You will most likely need to restart your service or role to have your configuration changes take effect.
Viewing and Editing Overridden Configuration Properties
To see a list of all role instances that have an override value for a particular configuration setting, navigate to
the entry for the configuration setting in the Status page, expand the Overridden by n instance(s) link in the
value cell for the overridden value.

16 | Cloudera Administration

Managing CDH and Managed Services

To view the override values, and change them if appropriate, click the Edit Overrides link. This opens the Edit
Overrides page, and lists the role instances that have override properties for the selected configuration setting.

On the Edit Overrides page, you can do any of the following:


View the list of role instances that have overridden the value specified in the role group. Use the selections
on the drop-down menu below the Value column header to view a list of instances that use the inherited
value, instances that use an override value, or all instances. This view is especially useful for finding
inconsistent properties in a cluster. You can also use the Host and Rack text boxes to filter the list.
Change the override value for the role instances to the inherited value from the associated role group. To do
so, select the role instances you want to change, choose Inherited Value from the drop-down menu next to
Change value of selected instances to and click Apply.
Change the override value for the role instances to a different value. To do so, select the role instances you
want to change, choose Other from the drop-down menu next to Change value of selected instances to.
Enter the new value in the text box and then click Apply.
Resetting Configuration Properties to the Default Value
To reset a property back to its default value, click the Reset to the default value link below the text box in the
value cell. The default value is inserted and both the text box and the Reset link disappear. Explicitly setting a
configuration to the same value as its default (inherited value) has the same effect as using the Reset to the
default value link.
There is no mechanism for resetting to an autoconfigured value. However, you can use the configuration history
and rollback feature to revert any configuration changes.
Restarting Services and Instances after Configuration Changes
If you change the configuration properties after you start a service or instance, you may need to restart the
service or instance to have the configuration properties become active. If you change configuration properties
at the service level that affect a particular role only (such as all DataNodes but not the NameNodes), you can
restart only that role; you do not need to restart the entire service. If you changed the configuration for a particular
role instance (such as one of four DataNodes), you may need to restart only that instance.
1. Follow the instructions in Restarting a Service on page 41 or Starting, Stopping, and Restarting Role Instances
on page 46.
2. If you see a Finished status, the service or role instances have restarted.

Cloudera Administration | 17

Managing CDH and Managed Services


3. Navigate to the Home > Status tab. The service should show a Status of Started for all instances and a health
status of Good.
For further information, see Stale Configurations on page 28.

Autoconfiguration
Cloudera Manager provides several interactive wizards to automate common workflows:

Installation - used to bootstrap a Cloudera Manager deployment


Add Cluster - used when adding a new cluster
Add Service - used when adding a new service
Upgrade - used when upgrading to a new version of Cloudera Manager
Static Service Pools - used when configuring static service pools
Import MapReduce - used when migrating from MapReduce to YARN

In some of these wizards, Cloudera Manager uses a set of rules to automatically configure certain settings to
best suit the characteristics of the deployment. For example, the number of hosts in the deployment drives the
memory requirements for certain monitoring daemons: the more hosts, the more memory is needed. Additionally,
wizards that are tasked with creating new roles will use a similar set of rules to determine an ideal host placement
for those roles.
Scope
The following table shows, for each wizard, the scope of entities it affects during autoconfiguration and role-host
placement.
Wizard

Autoconfiguration Scope

Role-Host Placement Scope

Installation

New cluster, Cloudera Management New cluster, Cloudera Management


Service
Service

Add Cluster

New cluster

New cluster

Add Service

New service

New service

Upgrade

Cloudera Management Service

Cloudera Management Service

Static Service Pools

Existing cluster

N/A

Import MapReduce

Existing YARN service

N/A

Certain autoconfiguration rules are unscoped, that is, they configure settings belonging to entities that aren't
necessarily the entities under the wizard's scope. These exceptions are explicitly listed.
Autoconfiguration
Cloudera Manager employs several different rules to drive automatic configuration, with some variation from
wizard to wizard. These rules range from the simple to the complex.

Configuration Scope
One of the points of complexity in autoconfiguration is configuration scope. The configuration hierarchy as it
applies to services is as follows: configurations may be modified at the service level (affecting every role in the
service), role group level (affecting every role instance in the group), or role level (affecting one role instance). A
configuration found in a lower level takes precedence over a configuration found in a higher level.
With the exception of the Static Service Pools, and the Import MapReduce wizard, all Cloudera Manager wizards
follow a basic pattern:
1. Every role in scope is moved into its own, new, role group.
2. This role group is the receptacle for the role's "idealized" configuration. Much of this configuration is driven
by properties of the role's host, which can vary from role to role.

18 | Cloudera Administration

Managing CDH and Managed Services


3. Once autoconfiguration is complete, new role groups with common configurations are merged.
4. The end result is a smaller set of role groups, each with an "idealized" configuration for some subset of the
roles in scope. A subset can have any number of roles; perhaps all of them, perhaps just one, and so on.
The Static Service Pools and Import MapReduce wizards configure role groups directly and do not perform any
merging.

Static Service Pools


Certain rules are only invoked in the context of the Static Service Pools wizard. Additionally, the wizard
autoconfigures cgroup settings for certain kinds of roles:

HDFS DataNodes
HBase RegionServers
MapReduce TaskTrackers
YARN NodeManagers
Impala Daemons
Solr Servers
Spark Workers
Accumulo Tablet Servers
Add-on services

YARN
yarn.nodemanager.resource.cpu-vcores - For each NodeManager role group, set to ((number of cores,

including hyperthreads, on one NodeManager member's host) * (service percentage chosen in wizard)).
All Services
Cgroup cpu.shares - For each role group that supports cpu.shares, set to max(20, (service percentage chosen
in wizard) * 20).
Cgroup blkio.weight - For each role group that supports blkio.weight, set to max(100, (service percentage
chosen in wizard) * 10).

Data Directories
Several autoconfiguration rules work with data directories, and there's a common sub-rule used by all such rules
to determine, out of all the mountpoints present on a host, which are appropriate for data. The subrule works
as follows:
The initial set of mountpoints for a host includes all those that are disk-backed. Network-backed mountpoints
are excluded.
Mountpoints beginning with /boot, /cdrom, /usr, /tmp, /home, or /dev are excluded.
Mountpoints beginning with /media are excluded, unless the backing device's name contains /xvd somewhere
in it.
Mountpoints beginning with /var are excluded, unless they are /var or /var/lib.
The largest mount point (in terms of total space, not available space) is determined.
Other mountpoints with less than 1% total space of the largest are excluded.
Mountpoints beginning with /var or equal to / are excluded unless theyre the largest mount point.
Remaining mountpoints are sorted lexicographically and retained for future use.

Memory
The rules used to autoconfigure memory reservations are perhaps the most complicated rules employed by
Cloudera Manager. When configuring memory, Cloudera Manager must take into consideration which roles are
likely to enjoy more memory, and must not over commit hosts if at all possible. To that end, it needs to consider
each host as an entire unit, partitioning its available RAM into segments, one segment for each role. To make
matters worse, some roles have more than one memory segment. For example, a Solr server has two memory

Cloudera Administration | 19

Managing CDH and Managed Services


segments: a JVM heap used for most memory allocation, and a JVM direct memory pool used for HDFS block
caching. Here is the overall flow during memory autoconfiguration:
1. The set of participants includes every host under scope as well as every {role, memory segment} pair on
those hosts. Some roles are under scope while others are not.
2. For each {role, segment} pair where the role is under scope, a rule is run to determine four different values
for that pair:
Minimum memory configuration. Cloudera Manager must satisfy this minimum, possibly over-committing
the host if necessary.
Minimum memory consumption. Like the above, but possibly scaled to account for inherent overhead.
For example, JVM memory values are multiplied by 1.3 to arrive at their consumption value.
Ideal memory configuration. If RAM permits, Cloudera Manager will provide the pair with all of this memory.
Ideal memory consumption. Like the above, but scaled if necessary.
3. For each {role, segment} pair where the role is not under scope, a rule is run to determine that pair's existing
memory consumption. Cloudera Manager will not configure this segment but will take it into consideration
by setting the pair's "minimum" and "ideal" to the memory consumption value.
4. For each host, the following steps are taken:
a. 20% of the host's available RAM is subtracted and reserved for the OS.
b. sum(minimum_consumption) and sum(ideal_consumption) are calculated.
c. An "availability ratio" is built by comparing the two sums against the host's available RAM.
a. If RAM < sum(minimum) ratio = 0
b. If RAM >= sum(ideal) ratio = 1
c. Otherwise, ratio is computed via: (RAM - sum(minimum)) / (sum(ideal) - sum(minimum))
5. For each {role, segment} pair where the role is under scope, the segment is configured to be (minimum +
((ideal - minimum) * (host availability ratio))). The value is rounded down to the nearest megabyte.
6. The {role, segment} pair is set with the value from the previous step. In the Static Service Pools wizard, the
role group is set just once (as opposed to each role).
7. Custom post-configuration rules are run.
Customization rules are applied in steps 2, 3 and 7. In step 2, there's a generic rule for most cases, as well as a
series of custom rules for certain {role, segment} pairs. Likewise, there's a generic rule to calculate memory
consumption in step 3 as well as some custom consumption functions for certain {role, segment} pairs.
Step 2 Generic Rule Excluding Static Service Pools Wizard
For every {role, segment} pair where the segment defines a default value, the pair's minimum is set to the
segment's minimum value (or 0 if undefined), and the ideal is set to the segment's default value.
Step 2 Custom Rules Excluding Static Service Pools Wizard
HDFS
For the NameNode and Secondary NameNode JVM heaps, the minimum is 50 MB and the ideal is max(1 GB,
sum_over_all(DataNode mountpoints available space) / 0.000008).
MapReduce
For the JobTracker JVM heap, the minimum is 50 MB and the ideal is max(1 GB, round((1 GB * 2.3717181092 *
ln(number of TaskTrackers in MapReduce service)) - 2.6019933306)). If there are <=5 TaskTrackers, the ideal is
1 GB.
For the mapper JVM heaps, the minimum is 1 and the ideal is (number of cores, including hyperthreads, on the
TaskTracker host). Note that memory consumption is scaled by mapred_child_java_opts_max_heap (the size
of a given task's heap).

20 | Cloudera Administration

Managing CDH and Managed Services


For the reducer JVM heaps, the minimum is 1 and the ideal is (number of cores, including hyperthreads, on the
TaskTracker host) / 2. Note that memory consumption is scaled by mapred_child_java_opts_max_heap (the
size of a given task's heap).
YARN
For the memory total allowed for containers, the minimum is 1 GB and the ideal is min(8 GB, (total RAM on
NodeManager host) * 0.8).
Hue
With the exception of the Beeswax Server (present only in CDH 4), Hue roles dont have memory limits. Therefore,
Cloudera Manager treats them as roles that consume a fixed amount of memory by setting their minimum and
ideal consumption values, but not their configuration values. The two consumption values are set to 256 MB.
Impala
With the exception of the Impala Daemon, Impala roles dont have memory limits. Therefore Cloudera Manager
treats them as roles that consume a fixed amount of memory by setting their minimum/ideal consumption
values, but not their configuration values. The two consumption values are set to 150 MB for the Catalog Server
and 64 MB for the StateStore.
For the Impala Daemon memory limit, the minimum is 256 MB and the ideal is ((total RAM on daemon host) *
0.64).
Solr
For the Solr Server JVM heap, the minimum is 50 MB and the ideal is (min(64 GB, (total RAM on Solr Server host)
* 0.64) / 2.6). For the Solr Server JVM direct memory segment, the minimum is 256 MB and the ideal is (min(64
GB, (total RAM on Solr Server host) * 0.64) / 2).
Cloudera Management Service
Alert Publisher JVM heap - treated as if it consumed a fixed amount of memory by setting the minimum/ideal
consumption values, but not the configuration values. The two consumption values are set to 256 MB.
Service and Host Monitor JVM heaps - the minimum is 50 MB and the ideal is either 256 MB (10 or fewer
managed hosts), 1 GB (100 or fewer managed hosts), or 2 GB (over 100 managed hosts).
Event Server, Reports Manager, and Navigator Audit Server JVM heaps - the minimum is 50 MB and the ideal
is 1 GB.
Navigator Metadata Server JVM heap - the minimum is 512 MB and the ideal is 2 GB.
Service and Host Monitor off-heap memory segments - the minimum is either 768 MB (10 or fewer managed
hosts), 2 GB (100 or fewer managed hosts), or 6 GB (over 100 managed hosts). The ideal is always twice the
minimum.
Step 2 Generic Rule for Static Service Pools Wizard
For every {role, segment} pair where the segment defines a default value and an autoconfiguration share, the
pair's minimum is set to the segment's default value, and the ideal is set to min((segment soft max (if exists) or
segment max (if exists) or 2^63-1), (total RAM on role's host * 0.8 / segment scale factor * service percentage
chosen in wizard * segment autoconfiguration share)).
Autoconfiguration shares are defined as follows:

HBase RegionServer JVM heap: 1


HDFS DataNode JVM heap: 1 in CDH 4, 0.2 in CDH 5
HDFS DataNode maximum locked memory: 0.8 (CDH 5 only)
Solr Server JVM heap: 0.5
Solr Server JVM direct memory: 0.5
Spark Worker JVM heap: 1
Cloudera Administration | 21

Managing CDH and Managed Services


Accumulo Tablet Server JVM heap: 1
Add-on services: any
Roles not mentioned here do not define autoconfiguration shares and thus aren't affected by this rule.
Additionally, there's a generic rule to handle cgroup.memory_limit_in_bytes, which is unused by Cloudera
services but is available for add-on services. Its behavior varies depending on whether the role in question has
segments or not.
With Segments
The minimum is the min(cgroup.memory_limit_in_bytes_min (if exists) or 0, sum_over_all(segment minimum
consumption)), and the ideal is the sum of all segment ideal consumptions.
Without Segments
The minimum is cgroup.memory_limit_in_bytes_min (if exists) or 0, and the ideal is (total RAM on role's host
* 0.8 * service percentage chosen in wizard).
Step 3 Custom Rules for Static Service Pools Wizard
YARN
For the memory total allowed for containers, the minimum is 1 GB and the ideal is min(8 GB, (total RAM on
NodeManager host) * 0.8 * service percentage chosen in wizard).
Impala
For the Impala Daemon memory limit, the minimum is 256 MB and the ideal is ((total RAM on Daemon host) *
0.8 * service percentage chosen in wizard).
MapReduce
Mapper JVM heaps - the minimum is 1 and the ideal is (number of cores, including hyperthreads, on the
TaskTracker host * service percentage chosen in wizard). Note that memory consumption is scaled by
mapred_child_java_opts_max_heap (the size of a given task's heap).
Reducer JVM heaps - the minimum is 1 and the ideal is (number of cores, including hyperthreads on the
TaskTracker host * service percentage chosen in wizard) / 2. Note that memory consumption is scaled by
mapred_child_java_opts_max_heap (the size of a given task's heap).
Step 3 Generic Rule
For every {role, segment} pair, the segment's current value is converted into bytes, and then multiplied by the
scale factor (1.0 by default, 1.3 for JVM heaps, and freely defined for Custom Service Descriptor services).
Step 3 Custom Rules
Impala
For the Impala Daemon, the memory consumption is 0 if YARN_For_ResourceManager is set. If the memory
limit is defined but not -1, its value is used verbatim. If it's defined but -1, the consumption is equal to the total
RAM on the Daemon host. If it is undefined, the consumption is (total RAM * 0.8).
MapReduce
See Step 3 Custom Rules for Static Service Pools Wizard on page 22.

22 | Cloudera Administration

Managing CDH and Managed Services


Solr
For the Solr Server JVM direct memory segment, the consumption is equal to the value verbatim provided
solr.hdfs.blockcache.enable and solr.hdfs.blockcache.direct.memory.allocation are both true.
Otherwise, the consumption is 0.
Step 7 Custom Rules
HDFS
NameNode JVM heaps are equalized. For every pair of NameNodes in an HDFS service with different heap
sizes, the larger heap size is reset to the smaller one.
JournalNode JVM heaps are equalized. For every pair of JournalNodes in an HDFS service with different heap
sizes, the larger heap size is reset to the smaller one.
NameNode and Secondary NameNode JVM heaps are equalized. For every {NameNode, Secondary NameNode}
pair in an HDFS service with different heap sizes, the larger heap size is reset to the smaller one.
HBase
Master JVM heaps are equalized. For every pair of Masters in an HBase service with different heap sizes, the
larger heap size is reset to the smaller one.
MapReduce
JobTracker JVM heaps are equalized. For every pair of JobTrackers in an MapReduce service with different heap
sizes, the larger heap size is reset to the smaller one.
Oozie
Oozie Server JVM heaps are equalized. For every pair of Oozie Servers in an Oozie service with different heap
sizes, the larger heap size is reset to the smaller one.
YARN
ResourceManager JVM heaps are equalized. For every pair of ResourceManagers in a YARN service with different
heap sizes, the larger heap size is reset to the smaller one.
ZooKeeper
ZooKeeper Server JVM heaps are equalized. For every pair of servers in a ZooKeeper service with different heap
sizes, the larger heap size is reset to the smaller one.
Impala
If an Impala service has YARN_For_ResourceManager set, every Impala Daemon memory limit is set to the
value of (yarn.nodemanager.resource.memory-mb * 1 GB) if there's a YARN NodeManager co-located with
the Impala Daemon.

General Rules
HDFS
dfs.datanode.du.reserved - For each DataNode, set to min((total space of DataNode host largest
mountpoint) / 10, 10 GB).
dfs.namenode.name.dir - For each NameNode, set to the first two mountpoints on the NameNode host
with /dfs/nn appended.
dfs.namenode.checkpoint.dir - For each Secondary NameNode, set to the first mountpoint on the
Secondary NameNode host with /dfs/snn appended.

Cloudera Administration | 23

Managing CDH and Managed Services


dfs.datanode.data.dir - For each DataNode, set to all the mountpoints on the host with /dfs/dn
appended.
dfs.journalnode.edits.dir - For each JournalNode, set to the first mountpoint on the JournalNode host
with /dfs/jn appended.
dfs.datanode.failed.volumes.tolerated - For each DataNode, set to (number of mountpoints on
DataNode host) / 2.
dfs.namenode.service.handler.count and dfs.namenode.handler.count - For each NameNode, set
to max(30, ln(number of DataNodes in this HDFS service) * 20).
dfs.block.local-path-access.user - For each HDFS service, set to impala if there's an Impala service
in the cluster. This rule is unscoped; it can fire even if the HDFS service is not under scope.
dfs.datanode.hdfs-blocks-metadata.enabled - For each HDFS service, set to true if there's an Impala
service in the cluster. This rule is unscoped; it can fire even if the HDFS service is not under scope.
dfs.client.read.shortcircuit - For each HDFS service, set to true if there's an Impala service in the
cluster. This rule is unscoped; it can fire even if the HDFS service is not under scope.
dfs.datanode.data.dir.perm - For each DataNode, set to 755 if there's an Impala service in the cluster
and the cluster isnt Kerberized. This rule is unscoped; it can fire even if the HDFS service is not under scope.
fs.trash.interval - For each HDFS service, set to 1.
MapReduce
mapred.local.dir - For each JobTracker, set to the first mountpoint on the JobTracker host with /mapred/jt
appended.
mapred.local.dir - For each TaskTracker, set to all the mountpoints on the host with /mapred/local
appended.
mapred.reduce.tasks - For each MapReduce service, set to max(1, sum_over_all(TaskTracker number of
reduce tasks (determined via mapred.tasktracker.reduce.tasks.maximum for that TaskTracker, which
is configured separately)) / 2).
mapred.job.tracker.handler.count - For each JobTracker, set to max(10, ln(number of TaskTrackers in
this MapReduce service) * 20).
mapred.submit.replication - If there's an HDFS service in the cluster, for each MapReduce service, set
to max(1, sqrt(number of DataNodes in the HDFS service)).
mapred.tasktracker.instrumentation - If there's a management service, for each MapReduce service,
set to org.apache.hadoop.mapred.TaskTrackerCmonInst. This rule is unscoped; it can fire even if the
MapReduce service is not under scope.
HBase
hbase.replication - For each HBase service, set to true if there's a Key-Value Store Indexer service in the
cluster. This rule is unscoped; it can fire even if the HBase service is not under scope.
replication.replicationsource.implementation - For each HBase service, set to
com.ngdata.sep.impl.SepReplicationSource if there's a Keystore Indexer service in the cluster. This
rule is unscoped; it can fire even if the HBase service is not under scope.
YARN
yarn.nodemanager.local-dirs - For each NodeManager, set to all the mountpoints on the NodeManager
host with /yarn/nm appended.
yarn.nodemanager.resource.cpu-vcores - For each NodeManager, set to the number of cores (including
hyperthreads) on the NodeManager host.
mapred.reduce.tasks - For each YARN service, set to max(1, sum_over_all(NodeManager number of
cores, including hyperthreads) / 2).
yarn.resourcemanager.nodemanagers.heartbeat-interval-ms - For each NodeManager, set to max(100,
10 * (number of NodeManagers in this YARN service)).

24 | Cloudera Administration

Managing CDH and Managed Services


yarn.scheduler.maximum-allocation-vcores - For each ResourceManager, set to
max_over_all(NodeManager number of vcores (determined via yarn.nodemanager.resource.cpu-vcores
for that NodeManager, which is configured separately)).
yarn.scheduler.maximum-allocation-mb - For each ResourceManager, set to max_over_all(NodeManager
amount of RAM (determined via yarn.nodemanager.resource.memory-mb for that NodeManager, which
is configured separately)).
mapreduce.client.submit.file.replication - If there's an HDFS service in the cluster, for each YARN
service, set to max(1, sqrt(number of DataNodes in the HDFS service)).
Hue
WebHDFS dependency - For each Hue service, set to either the first HttpFS role in the cluster, or, if there are
none, the first NameNode in the cluster.
HBase Thrift Server dependency- For each Hue service in a CDH 4.4 or later cluster, set to the first HBase
Thrift Server in the cluster.
Impala
For each Impala service, set Enable Impala auditing for Navigator to true if there's a Cloudera Management
Service with a Navigator audit server role. This rule is unscoped; it can fire even if the Impala service is not under
scope.
All Services
If a service dependency is unset, and a service with the desired type exists in the cluster, set the service
dependency to the first such target service. Applies to all service dependencies except
YARN_For_ResourceManager. Applies only to the Express and Add Cluster wizards.
Role-Host Placement
Cloudera Manager employs the same role-host placement rule regardless of wizard. The set of hosts considered
depends on the scope. If the scope is a cluster, all hosts in the cluster are included. If a service, all hosts in the
service's cluster are included. If the Cloudera Management Service, all hosts in the deployment are included. The
rules are as follows:
1. The hosts are sorted from most to least physical RAM. Ties are broken by sorting on hostname (ascending)
followed by host identifier (ascending).
2. The overall number of hosts is used to determine which arrangement to use. These arrangements are
hard-coded, each dictating for a given "master" role type, what index (or indexes) into the sorted host list in
step 1 to use.
3. Master role types are included based on several factors:

Is this role type part of the service (or services) under scope?
Does the service already have the right number of instances of this role type?
Does the cluster's CDH version support this role type?
Does the installed Cloudera Manager license allow for this role type to exist?

4. Master roles are placed on each host using the indexes and the sorted host list. If a host already has a given
master role, it is skipped.
5. An HDFS DataNode is placed on every host outside of the arrangement described in step 2, provided HDFS
is one of the services under scope.
6. Certain "worker" roles are placed on every host where an HDFS DataNode exists, either because it existed
there prior to the wizard, or because it was added in the previous step. The supported worker role types are:

MapReduce TaskTrackers
YARN NodeManagers
HBase RegionServers
Impala Daemons
Cloudera Administration | 25

Managing CDH and Managed Services


Spark Workers
7. Hive gateways are placed on every host, provided a Hive service is under scope and a gateway didnt already
exist on a given host.
8. Spark on YARN gateways are placed on every host, provided a Spark on YARN service is under scope and a
gateway didnt already exist on a given host.
This rule merely dictates the default placement of roles; you are free to modify it before it is applied by the
wizard.

Custom Configuration
Required Role:
Cloudera Manager exposes properties that allow you to insert custom configuration text into XML configuration,
property, and text files, or into an environment. The naming convention for these properties is: XXX Advanced
Configuration Snippet (Safety Valve) for YYY or XXX YYY Advanced Configuration Snippet (Safety Valve), where
XXX is a service or role and YYY is the target.
The values you enter into a configuration snippet must conform to the syntax of the target. For an XML
configuration file, the configuration snippet must contain valid XML property definitions. For a properties file,
the configuration snippet must contain valid property definitions. Some files simply require a list of host
addresses.
The configuration snippet mechanism is intended for use in cases where there is configuration setting that is
not exposed as a configuration property in Cloudera Manager. Configuration snippets generally override normal
configuration. Contact Cloudera Support if you are required to use a configuration snippet that is not explicitly
documented.
Service-wide configuration snippets apply to all roles in the service; a configuration snippet for a role group
applies to all instances of the role associated with that role group.
There are configuration snippets for servers and client configurations. In general after changing a server
configuration snippet you must restart the server, and after changing a client configuration snippet you must
redeploy the client configuration. Sometimes you can refresh instead of restart. In some cases however, you
must restart a dependent server after changing a client configuration. For example, changing a MapReduce
client configuration marks the dependent Hive server as stale, which must be restarted. The Admin Console
displays an indicator when a server must be restarted. In addition, the All Configuration Issues tab on the Home
page lists the actions you must perform to propagate the changes.
Configuration Snippet Types and Syntax
Type

Description

Syntax

Configuration

Set configuration properties in


various configuration files; the
property name indicates into which
configuration file the configuration
will be placed. Configuration files
have the extension .xml or .conf.

<property>
<name>property_name</name>
<value>property_value</value>
</property>

For example, to specify a MySQL connector library, put


this property definition in that configuration snippet:
<property>

<name>hive.aux.jars.path</name>
For example, there are several
configuration snippets for the Hive
<value>file:///usr/share/java/mysql-connector-java.jar</value>
service. One Hive configuration
</property>
snippet property is called the
HiveServer2 Advanced Configuration
Snippet for hive-site.xml;
configuration you enter here is
inserted verbatim into the

26 | Cloudera Administration

Managing CDH and Managed Services


Type

Description

Syntax

hive-site.xml file associated with

the HiveServer2 role group.


To see a list of configuration
snippets that apply to a specific
configuration file, enter the
configuration file name in the Search
field in the top navigation bar. For
example, searching for
mapred-site.xml shows the
configuration snippets that have
mapred-site.xml in their name.
Environment

Specify key-value pairs for a service, key=value


role, or client that are inserted into For example, to add JDBC connectors to a Hive gateway
the respective environment.
classpath, add
One example of using an
AUX_CLASSPATH=/usr/share/java/mysql-connector-java.jar:\
environment configuration snippet /usr/share/java/oracle-connector-java.jar
is to add a JAR to a classpath. Place or
JARs in a custom location such as
AUX_CLASSPATH=/usr/share/java/*
/opt/myjars and extend the
classpath via the appropriate service to Gateway Client Advanced Configuration Snippet for
environment configuration snippet. hive-env.sh.
The value of a JAR property must
conform to the syntax supported by
its environment. See Setting the
class path.
Do not place JARs inside locations
such as /opt/cloudera or
/usr/lib/{hadoop*,hbase*,hive*,etc.}

that are managed by Cloudera


because they are overwritten at
upgrades.
Logging

Set log4j properties in a


log4j.properties file.

key1=value1
key2=value2

For example:
log4j.rootCategory=INFO, console
max.log.file.size=200MB
max.log.file.backup.index=10

Metrics

Set properties to configure Hadoop


metrics in a
hadoop-metrics.properties or
hadoop-metrics2.properties file.

key1=value1
key2=value2

For example:
*.sink.foo.class=org.apache.hadoop.metrics2.sink.FileSink
namenode.sink.foo.filename=/tmp/namenode-metrics.out
secondarynamenode.sink.foo.filename=/tmp/secondarynamenode-metrics.out

White and black Specify a list of host addresses that host1.domain1


host2.domain2
lists
are allowed or disallowed from
accessing a service.
Setting an Advanced Configuration Snippet
1. Click a service.

Cloudera Administration | 27

Managing CDH and Managed Services


2.
3.
4.
5.
6.
7.

Click the Configuration tab.


In the Search box, type Advanced Configuration Snippet.
Choose a property that contains the string Advanced Configuration Snippet (Safety Valve).
Specify the snippet properties.
Click Save Changes to commit the changes.
Restart the service or role or redeploy client configurations as indicated.

Setting Advanced Configuration Snippets for a Cluster or Clusters


1. Do one of the following
specific cluster
1. On the Home > Status tab, click a cluster name.
2. Select Configuration > Advanced Configuration Snippets.
all clusters
1. Select Configuration > Advanced Configuration Snippets.
2. Specify the snippet properties.
3. Click Save Changes to commit the changes.
4. Restart the service or role or redeploy client configurations as indicated.

Stale Configurations
Required Role:
The Stale Configurations page provides differential views of changes made in a cluster. For any configuration
change, the page contains entries of all affected attributes. For example, the following File entry shows the
change to the file hdfs-site.xml when you update the property controlling how much disk space is reserved
for non-HDFS use on each DataNode:

To display the entities affected by a change, click the Show button at the right of the entry. The following dialog
shows that three DataNodes were affected by the disk space change:

28 | Cloudera Administration

Managing CDH and Managed Services


Viewing Stale Configurations
To view stale configurations, click the , , or indicator next to a service on the Cloudera Manager Admin
Console Home Page or on a service status page.
Attribute Categories
The categories of attributes include:
Environment - represents environment variables set for the role. For example, the following entry shows
the change to the environment that occurs when you update the heap memory configuration of the
SecondaryNameNode.

Files - represents configuration files used by the role.


Process User & Group - represents the user and group for the role. Every role type has a configuration to
specify the user/group for the process. If you change a value for a user or group on any service's configuration
page it will appear in the Stale Configurations page.
System Resources - represents system resources allocated for the role, including ports, directories, and
cgroup limits. For example, a change to the port of role instance will appear in the System Resources category.
Client Configs Metadata - represents client configurations.
Filtering Stale Configurations
You filter the entries on the Stale Configurations page by selecting from one of the drop-down lists:
Attribute - you can filter by an attribute category such as All Files or by a specific file such as topology.map
or yarn-site.xml.
Service
Role
After you make a selection, both the page and the drop-down show only entries that match that selection.
To reset the view, click Remove Filter or select All XXX, where XXX is Files, Services, or Roles, from the drop-down.
For example, to see all the files, select All Files.
Stale Configuration Actions
The Stale Configurations page displays action buttons. The action depends on what is required to bring the
entire cluster's configuration up to date. If you navigate to the page by clicking a (Refresh Needed) indicator,
the action button will say Restart Stale Services if one of the roles listed on the page need to be restarted.

Refresh Stale Services - Refreshes stale services.


Restart Stale Services - Restarts stale services.
Restart Cloudera Management Service - Runs the restart Cloudera Management Service action.
Deploy Client Configuration - Runs the cluster deploy client configurations action.

Client Configuration Files


Required Role:
To allow clients to use the HBase, HDFS, Hive, MapReduce, and YARN services, Cloudera Manager creates zip
archives of the configuration files containing the service properties. The zip archive is referred to as a client
configuration file. Each archive contains the set of configuration files needed to access the service: for example,

Cloudera Administration | 29

Managing CDH and Managed Services


the MapReduce client configuration file contains copies of core-site.xml, hadoop-env.sh, hdfs-site.xml,
log4j.properties, and mapred-site.xml.
Client configuration files are generated automatically by Cloudera Manager based on the services and roles you
have installed and Cloudera Manager deploys these configurations automatically when you install your cluster,
add a service on a host, or add a gateway role on a host. Specifically, for each host that has a service role instance
installed, and for each host that is configured as a gateway role for that service, the deploy function downloads
the configuration zip file, unzips it into the appropriate configuration directory, and uses the Linux alternatives
mechanism to set a given, configurable priority level. If you are installing on a system that happens to have
pre-existing alternatives, then it is possible another alternative may have higher priority and will continue to
be used. The alternatives priority of the Cloudera Manager client configuration is configurable under the Gateway
scope of the Configuration tab for the appropriate service.
You can also manually distribute client configuration files to the clients of a service.
The main circumstance that may require a redeployment of the client configuration files is when you have
modified a configuration. In this case you will typically see a message instructing you to redeploy your client
configurations. The affected service(s) will also display a icon. Click the indicator to display the Stale
Configurations on page 28 page.
How Client Configurations are Deployed
Client configuration files are deployed on any host that is a client for a servicethat is, that has a role for the
service on that host. This includes roles such as DataNodes, TaskTrackers, RegionServers and so on as well as
gateway roles for the service.
If roles for multiple services are running on the same host (for example, a DataNode role and a TaskTracker role
on the same host) then the client configurations for both roles are deployed on that host, with the alternatives
priority determining which configuration takes precedence.
For example, suppose we have six hosts running roles as follows: host H1: HDFS-NameNode; host H2:
MR-JobTracker; host H3: HBase-Master; host H4: MR-TaskTracker, HDFS-DataNode, HBase-RegionServer; host
H5: MR-Gateway; host H6: HBase-Gateway. Client configuration files will be deployed on these hosts as follows:
host H1: hdfs-clientconfig (only); host H2: mapreduce-clientconfig, host H3: hbase-clientconfig; host H4:
hdfs-clientconfig, mapreduce-clientconfig, hbase-clientconfig; host H5: mapreduce-clientconfig; host H6:
hbase-clientconfig
If the HDFS NameNode and MapReduce JobTracker were on the same host, then that host would have both
hdfs-clientconfig and mapreduce-clientconfig installed.
Downloading Client Configuration Files
1. Follow the appropriate procedure according to your starting point:
Page

Procedure

Home

1.

Service

1. Go to a service whose client configuration you want to download.


2. Select Actions > Download Client Configuration.

On the Home > Status tab, click


to the right of the cluster name and select
View Client Configuration URLs. A pop-up window with links to the
configuration files for the services you have installed displays.
2. Click a link or save the link URL and download the file using wget or curl.

Manually Redeploying Client Configuration Files


Although Cloudera Manager will deploy client configuration files automatically in many cases, if you have modified
the configurations for a service, you may need to redeploy those configuration files.
If your client configurations were deployed automatically, the command described in this section will attempt
to redeploy them as appropriate.

30 | Cloudera Administration

Managing CDH and Managed Services


Note: If you are deploying client configurations on a host that has multiple services installed, some
of the same configuration files, though with different configurations, will be installed in the conf
directories for each service. Cloudera Manager uses the priority parameter in the alternatives
--install command to ensure that the correct configuration directory is made active based on the
combination of services on that host. The priority order is YARN > MapReduce > HDFS. The priority
can be configured under the Gateway sections of the Configuration tab for the appropriate service.
1.

On the Home > Status tab, click


to the right of the cluster name and select Deploy Client Configuration.
2. Click Deploy Client Configuration.

Viewing and Reverting Configuration Changes


Required Role:
Important: This feature is available only with a Cloudera Enterprise license; it is not available in
Cloudera Express. For information on Cloudera Enterprise licenses, see Managing Licenses on page
447.
Whenever you change and save a set of configuration settings for a service or role instance or a host, Cloudera
Manager saves a revision of the previous settings and the name of the user who made the changes. You can
then view past revisions of the configuration settings, and, if desired, roll back the settings to a previous state.
Viewing Configuration Changes
1. For a service, role, or host, and click the Configuration tab and the History and Rollback button. The most
recent revision, currently in effect, is shown under Current Revision. Prior revisions are shown under Past
Revisions.
By default, or if you click Show All, a list of all revisions is shown. If you are viewing a service or role
instance, all service/role group related revisions are shown. If you are viewing a host or all hosts, all
host/all hosts related revisions are shown.
To list only the configuration revisions that were done in a particular time period, use the Time Range
Selector to select a time range. Then, click Show within the Selected Time Range.
2. Click the Details... link. The Revision Details dialog displays.
Revision Details Dialog
For a service or role instance, shows the following:

A brief message describing the context of the changes


The date/time stamp of the change
The user who performed the change
The names of any role groups created
The names of any role groups deleted

For a host instance, shows just a message, date and time stamp, and the user.
The dialog contains two tabs:
Configuration Values - displays configuration value changes, where changes are organized under the role
group to which they were applied. (For example, if you changed a Service-Wide property, it will affect all role
groups for that service). For each modified property, the Value column shows the new value of the property
and the previous value.
Group Membership - displays changes to the changed the group membership of a role instance (moved the
instance from one group to another). This tab is only shown for service and role configurations.

Cloudera Administration | 31

Managing CDH and Managed Services


Reverting Configuration Changes
1.
2.
3.
4.

Select the current or past revision to which to roll back.


Click the Details... link. The Revision Details dialog displays.
Click the Configuration Values tab.
Click the Revert Configuration Changes button. The revert action occurs immediately. You may need to restart
the service or the affected roles for the change to take effect.
Important: This feature can only be used to revert changes to configuration values. You cannot use
this feature to:
Revert NameNode high availability. You must perform this action by explicitly disabling high
availability.
Disable Kerberos security.
Revert role group actions (creating, deleting, or moving membership among groups). You must
perform these actions explicitly in the Role Groups on page 47 feature.

Exporting and Importing Cloudera Manager Configuration


You can use the Cloudera Manager API to programmatically export and import a definition of all the entities in
your Cloudera Manager-managed deploymentclusters, service, roles, hosts, users and so on. See the Cloudera
Manager API documentation on how to manage deployments using the /cm/deployment resource.

Managing Clusters
Cloudera Manager can manage multiple clusters, however each cluster can only be associated with a single
Cloudera Manager Server or Cloudera Manager HA pair. Once you have successfully installed your first cluster,
you can add additional clusters, running the same or a different version of CDH. You can then manage each
cluster and its services independently.
On the Home > Status tab you can access many cluster-wide actions by selecting
to the right of the cluster
name: add a service, start, stop, restart, deploy client configurations, enable Kerberos, and perform cluster refresh,
rename, upgrade, and maintenance mode actions.
Note:
Cloudera Manager configuration screens offer two layout options: classic and new. The new layout
is the default; however, on each configuration page you can easily switch between layouts using the
Switch to XXX layout link at the top right of the page. For more information, see Configuration Overview
on page 8.

Adding and Deleting Clusters


Required Role:
Cloudera Manager can manage multiple clusters. Furthermore, the clusters do not need to run the same version
of CDH; you can manage both CDH 4 and CDH 5 clusters with Cloudera Manager.
Adding a Cluster
Action

Procedure

New Hosts

1.

32 | Cloudera Administration

On the Home > Status tab, click


and select Add Cluster. This begins the Installation
Wizard, just as if you were installing a cluster for the first time. (See Cloudera Manager
Deployment for detailed instructions.)

Managing CDH and Managed Services


Action

Procedure
2. To find new hosts, not currently managed by Cloudera Manager, where you want to install
CDH, enter the host names or IP addresses, and click Search. Cloudera Manager lists the
hosts you can use to configure a new cluster. Managed hosts that already have services
installed will not be selectable.
3. Click Continue to install the new cluster. At this point the installation continues through
the wizard the same as it did when you installed your first cluster. You will be asked to
select the version of CDH to install, which services you want and so on, just as previously.
4. Restart the Reports Manager role.

Managed Hosts

You may have hosts that are already "managed" but are not part of a cluster. You can have
managed hosts that are not part of a cluster when you have added hosts to Cloudera Manager
either through the Add Host wizard, or by manually installing the Cloudera Manager agent
onto hosts where you have not install any other services. This will also be the case if you
remove all services from a host so that it no longer is part of a cluster.
1.

On the Home > Status tab, click


and select Add Cluster. This begins the Installation
Wizard, just as if you were installing a cluster for the first time. (See Cloudera Manager
Deployment for detailed instructions.)
2. To see the list of the currently managed hosts, click the Currently Managed Hosts tab.
This tab does not appear if you have no currently managed hosts that are not part of a
cluster.
3. To perform the installation, click Continue. Instead of searching for hosts, this will attempt
to install onto any hosts managed by Cloudera Manager that are not already part of a
cluster. It will proceed with the installation wizard as for a new cluster installation.
4. Restart the Reports Manager role.
Deleting a Cluster
1. Stop the cluster.
2.
On the Home > Status tab, click

to the right of the cluster name and select Delete.

Starting, Stopping, Refreshing, and Restarting a Cluster


Required Role:
Starting a Cluster
1.

On the Home > Status tab, click


to the right of the cluster name and select Start.
2. Click Start that appears in the next screen to confirm. The Command Details window shows the progress of
starting services.
When All services successfully started appears, the task is complete and you can close the Command Details
window.
Note: The cluster-level Start action starts only CDH and other product services (Impala, Cloudera
Search). It does not start the Cloudera Management Service. You must start the Cloudera Management
Service separately if it is not already running.
Stopping a Cluster
1.

On the Home > Status tab, click


to the right of the cluster name and select Stop.
2. Click Stop in the confirmation screen. The Command Details window shows the progress of stopping services.
Cloudera Administration | 33

Managing CDH and Managed Services


When All services successfully stopped appears, the task is complete and you can close the Command Details
window.
Note: The cluster-level Stop action does not stop the Cloudera Management Service. You must stop
the Cloudera Management Service separately.
Refreshing a Cluster
Runs a cluster refresh action to bring the configuration up to date without restarting all services. For example,
certain masters (for example NameNode and ResourceManager) have some configuration files (for example,
fair-scheduler.xml, mapred_hosts_allow.txt, topology.map) that can be refreshed. If anything changes
in those files then a refresh can be used to update them in the master. Here is a summary of the operations
performed in a refresh action:

To refresh a cluster, in the Home > Status tab, click

to the right of the cluster name and select Refresh Cluster.

Restarting a Cluster
1.

On the Home > Status tab, click


to the right of the cluster name and select Restart.
2. Click Restart that appears in the next screen to confirm. The Command Details window shows the progress
of stopping services.
When All services successfully started appears, the task is complete and you can close the Command Details
window.

Renaming a Cluster
Required Role:
1.

On the Home > Status tab, click


to the right of the cluster name and select Rename Cluster.
2. Type the new cluster name and click Rename Cluster.

Cluster-Wide Configuration
Required Role:
To make configuration changes that apply to an entire cluster, do one of the following to open the configuration
page:
34 | Cloudera Administration

Managing CDH and Managed Services


all clusters
1. Select Configuration and then select one of the following classes of properties:

Advanced Configuration Snippets


Databases
Disk Space Thresholds
Local Data Directories
Local Data Files
Log Directories
Navigator Settings
Non-default Values - properties whose value differs from the default value
Non-uniform Values - properties whose values are not uniform across the cluster or clusters
Port Configurations
Service Dependencies

You can also select Configuration Issues to view a list of configuration issues for all clusters.
specific cluster
1. On the Home page, click a cluster name.
2. Select Configuration and then select one of the classes of properties listed above.
You can also apply the following filters to limit the displayed properties:
Enter a search term in the Search box to search for properties by name or description.
Expand the Status filter to select options that limit the displayed properties to those with errors or warnings,
properties that have been edited, properties with non-default values, or properties with overrides. Select All
to remove any filtering by Status.
Expand the Scope filter to display a list of service types. Expand a service type heading to filter on Service-Wide
configurations for a specific service instance or select one of the default role groups listed under each service
type. Select All to remove any filtering by Scope.
Expand the Category filter to filter using a sub-grouping of properties. Select All to remove any filtering by
Category.

Moving a Host Between Clusters


Required Role:
Moving a host between clusters can be accomplished by:
1. Decommissioning the host (see Decommissioning Role Instances on page 46).
2. Removing all roles from the host (except for the Cloudera Manager management roles). See Deleting Role
Instances on page 47.
3. Deleting the host from the cluster (see Deleting Hosts on page 61), specifically the section on removing a
host from a cluster but leaving it available to Cloudera Manager.
4. Adding the host to the new cluster (see Adding a Host to the Cluster on page 54).
5. Adding roles to the host (optionally using one of the host templates associated with the new cluster). See
Adding a Role Instance on page 45 and Host Templates on page 57.

Managing Services
Cloudera Manager service configuration features let you manage the deployment and configuration of CDH and
managed services. You can add new services and roles if needed, gracefully start, stop and restart services or
roles, and decommission and delete roles or services if necessary. Further, you can modify the configuration
properties for services or for individual role instances. If you have a Cloudera Enterprise license, you can view
past configuration changes and roll back to a previous revision. You can also generate client configuration files,
enabling you to easily distribute them to the users of a service.
Cloudera Administration | 35

Managing CDH and Managed Services


The topics in this chapter describe how to configure and use the services on your cluster. Some services have
unique configuration requirements or provide unique features: those are covered in Managing Individual Services
on page 77.
Note:
Cloudera Manager configuration screens offer two layout options: classic and new. The new layout
is the default; however, on each configuration page you can easily switch between layouts using the
Switch to XXX layout link at the top right of the page. For more information, see Configuration Overview
on page 8.

Adding a Service
Required Role:
After initial installation, you can use the Add a Service wizard to add and configure new service instances. For
example, you may want to add a service such as Oozie that you did not select in the wizard during the initial
installation.
The binaries for the following services are not packaged in CDH 4 and CDH 5 and must be installed individually
before being adding the service:
Service

Installation Documentation

Accumulo

Apache Accumulo Documentation

Impala (not in CDH 4)

Installing Impala

Search (not in CDH 4)

Installing Search

Kafka

Installing Kafka

Key Trustee KMS

Installing Key Trustee KMS

If you do not add the binaries before adding the service, the service will fail to start.
To add a service:
1.

On the Home > Status tab, click


to the right of the cluster name and select Add a Service. A list of service
types display. You can add one type of service at a time.
2. Click the radio button next to a service and click Continue. If you are missing required binaries, a pop-up
displays asking if you want to continue with adding the service.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

36 | Cloudera Administration

Managing CDH and Managed Services


Range Definition

Matching Hosts

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Review and modify configuration settings, such as data directory paths and heap sizes and click Continue.
The service is started.
6. Click Continue then click Finish. You are returned to the Home page.
7. Verify the new service is started properly by checking the health status for the new service. If the Health
Status is Good, then the service started properly.

Comparing Configurations for a Service Between Clusters


Required Role:
To compare the configuration settings for a particular service between two different clusters in a Cloudera
Manager deployment, perform the following steps:
1. On the Home > Status tab, click the name of the service you want to compare, or click the Clusters menu and
select the name of the service.
2. Click the Configuration tab.
3. Click the drop-down menu above the Filters pane, and select from one of the options that begins Diff with...:
service on cluster - For example, HBASE-1 on Cluster 1. This is the default display setting. All properties
are displayed for the selected instance of the service.
service on all clusters - For example, HBase on all clusters. All properties are displayed for all instances
of the service.
Diff with service on cluster - For example, Diff with HBase on Cluster 2. Properties are displayed only if
the values for the instance of the service whose page you are on differ from the values for the instance
selected in the drop-down menu.
Diff with service on all clusters - For example, Diff with HBase on all clusters. Properties are displayed if
the values for the instance of the service whose page you are on differ from the values for one or more
other instances in the Cloudera Manager deployment.
The service's properties will be displayed showing the values for each property for the selected clusters. The
filters on the left side can be used to limit the properties displayed.
You can also view property configuration values that differ between clusters across a deployment by selecting
Non-uniform Values on the Configuration tab of the Cloudera Manager Home > Status tab. For more information,
see Cluster-Wide Configuration on page 34

Add-on Services
Required Role:
Cloudera Manager supports adding new types of services (referred to as an add-on service) to Cloudera Manager,
allowing such services to leverage Cloudera Manager distribution, configuration, monitoring, resource management,
and life-cycle management features. An add-on service can be provided by Cloudera or an independent software
vendor (ISV). If you have multiple clusters managed by Cloudera Manager, an add-on service can be deployed
on any of the clusters.

Cloudera Administration | 37

Managing CDH and Managed Services


Note: If the add-on service is already installed and running on hosts that are not currently being
managed by Cloudera Manager, you must first add the hosts to a cluster that's under management.
See Adding a Host to the Cluster on page 54 for details.
Custom Service Descriptor Files
Integrating an add-on service requires a Custom Service Descriptor (CSD) file. A CSD file contains all the
configuration needed to describe and manage a new service. A CSD is provided in the form of a JAR file.
Depending on the service, the CSD and associated software may be provided by Cloudera or by an ISV. The
integration process assumes that the add-on service software (parcel or package) has been installed and is
present on the cluster. The recommended method is for the ISV to provide the software as a parcel, but the
actual mechanism for installing the software is up to the ISV. The instructions in Installing an Add-on Service
on page 38 assume that you have obtained the CSD file from the Cloudera repository or from an ISV. It also
assumes you have obtained the service software, ideally as a parcel, and have or will install it on your cluster
either prior to installing the CSD or as part of the CSD installation process.
Configuring the Location of Custom Service Descriptor Files
The default location for CSD files is /opt/cloudera/csd. You can change the location in the Cloudera Manager
Admin Console as follows:
1.
2.
3.
4.

Select Administration > Settings.


Click the Custom Service Descriptors category.
Edit the Local Descriptor Repository Path property.
Click Save Changes to commit the changes.

Installing an Add-on Service


An ISV may provide its software in the form or a parcel, or they may have a different way of installing their
software onto your cluster. If their installation process is not via a parcel, then you should install their software
before adding the CSD file. Follow the instructions from the ISV for installing the software, if you have not done
so already. If the ISV has provided their software as a parcel, they may also have included the location of their
parcel repository in the CSD they have provided. In that case, install the CSD first and then install the parcel.
Installing the Custom Service Descriptor File
1. Acquire the CSD file from Cloudera or an ISV.
2. Log on to the Cloudera Manager Server host, and place the CSD file under the location configured for CSD
files.
3. Set the file ownership to cloudera-scm:cloudera-scm with permission 644.
4. Restart the Cloudera Manager Server:
service cloudera-scm-server restart

5. Log into the Cloudera Manager Admin Console and restart the Cloudera Management Service.
a. Do one of the following:
1. Select Clusters > Cloudera Management Service > Cloudera Management Service.
2. Select Actions > Restart.

On the Home > Status tab, click

to the right of Cloudera Management Service and select Restart.

b. Click Restart to confirm. The Command Details window shows the progress of stopping and then starting
the roles.
c. When Command completed with n/n successful subcommands appears, the task is complete. Click Close.

38 | Cloudera Administration

Managing CDH and Managed Services


Installing the Parcel
If you have already installed the external software onto your cluster, you can skip these steps and proceed to
Adding an Add-on Service on page 39.
1. Click
in the main navigation bar. If the vendor has included the location of the repository in the CSD, the
parcel should already be present and ready for downloading. If the parcel is available, skip to step 7.
2. Use one of the following methods to open the parcel settings page:
Navigation bar
1. Click
in the top navigation bar or click Hosts and click the Parcels tab.
2. Click the Edit Settings button.
Menu
1. Select Administration > Settings.
2. Select Category > Parcels .
3. In the Remote Parcel Repository URLs list, click to open an additional row.
4. Enter the path to the repository.
5. Click Save Changes to commit the changes.
6. Click . The external parcel should appear in the set of parcels available for download.
7. Download, distribute, and activate the parcel. See Managing Parcels.
Adding an Add-on Service
Add the service following the procedure in Adding a Service on page 36.
Uninstalling an Add-on Service
1. Stop all instances of the service.
2. Delete the service from all clusters. If there are other services that depend on the service you are trying to
delete, you must delete those services first.
3. Log on to the Cloudera Manager Server host and remove the CSD file.
4. Restart the Cloudera Manager Server:
service cloudera-scm-server restart

5. After the server has restarted, log into the Cloudera Manager Admin Console and restart the Cloudera
Management Service.
6. Optionally remove the parcel.

Starting, Stopping, and Restarting Services


Required Role:
Starting and Stopping Services
It's important to start and stop services that have dependencies in the correct order. For example, because
MapReduce and YARN have a dependency on HDFS, you must start HDFS before starting MapReduce or YARN.
The Cloudera Management Service and Hue are the only two services on which no other services depend; although
you can start and stop them at anytime, their preferred order is shown in the following procedures.
The Cloudera Manager cluster actions start and stop services in the correct order. To start or stop all services
in a cluster, follow the instructions in Starting, Stopping, Refreshing, and Restarting a Cluster on page 33.

Cloudera Administration | 39

Managing CDH and Managed Services


Starting a Service on All Hosts
1.

On the Home > Status tab, click


to the right of the service name and select Start.
2. Click Start that appears in the next screen to confirm. When you see a Finished status, the service has started.
The order in which to start services is:
1. Cloudera Management Service
2. ZooKeeper
3. HDFS
4. Solr
5. Flume
6. HBase
7. Key-Value Store Indexer
8. MapReduce or YARN
9. Hive
10. Impala
11. Oozie
12. Sqoop
13. Hue
Note: If you are unable to start the HDFS service, it's possible that one of the roles instances, such
as a DataNode, was running on a host that is no longer connected to the Cloudera Manager Server
host, perhaps because of a hardware or network failure. If this is the case, the Cloudera Manager
Server will be unable to connect to the Cloudera Manager Agent on that disconnected host to start
the role instance, which will prevent the HDFS service from starting. To work around this, you can
stop all services, abort the pending command to start the role instance on the disconnected host,
and then restart all services again without that role instance. For information about aborting a pending
command, see Aborting a Pending Command on page 43.
Stopping a Service on All Hosts
1.

On the Home > Status tab, click


to the right of the service name and select Stop.
2. Click Stop that appears in the next screen to confirm. When you see a Finished status, the service has stopped.
The order in which to stop services is:
1. Hue
2. Sqoop
3. Oozie
4. Impala
5. Hive
6. MapReduce or YARN
7. Key-Value Store Indexer
8. HBase
9. Flume
10. Solr
11. HDFS
12. ZooKeeper
13. Cloudera Management Service

40 | Cloudera Administration

Managing CDH and Managed Services


Restarting a Service
It is sometimes necessary to restart a service, which is essentially a combination of stopping a service and then
starting it again. For example, if you change the hostname or port where the Cloudera Manager is running, or
you enable TLS security, you must restart the Cloudera Management Service to update the URL to the Server.
1.

On the Home > Status tab, click


to the right of the service name and select Restart.
2. Click Start on the next screen to confirm. When you see a Finished status, the service has restarted.
To restart all services, use the restart cluster action.

Rolling Restart
Required Role:
Important: This feature is available only with a Cloudera Enterprise license; it is not available in
Cloudera Express. For information on Cloudera Enterprise licenses, see Managing Licenses on page
447.
Rolling restart allows you to conditionally restart the role instances of Flume, HBase, HDFS, Kafka, MapReduce,
YARN, and ZooKeeper services to update software or use a new configuration. If the service is not running, rolling
restart is not available for that service. You can do a rolling restart of each service individually.
If you have HDFS high availability enabled, you can also perform a cluster-level rolling restart. At the cluster
level, the rolling restart of worker hosts is performed on a host-by-host basis, rather than per service, to avoid
all roles for a service potentially being unavailable at the same time. During a cluster restart, in order to avoid
having your NameNode (and thus the cluster) being unavailable during the restart, Cloudera Manager will force
a failover to the standby NameNode.
MapReduce (MRv1) JobTracker High Availability on page 324 and YARN (MRv2) ResourceManager High Availability
on page 315 is not required for a cluster-level rolling restart. However, if you have JobTracker or ResourceManager
high availability enabled, Cloudera Manager will force a failover to the standby JobTracker or ResourceManager.
Performing a Service or Role Rolling Restart
You can initiate a rolling restart from either the Status page for one of the eligible services, or from the service's
Instances page, where you can select individual roles to be restarted.
1. Go to the service you want to restart.
2. Do one of the following:
service - Select Actions > Rolling Restart.
role 1. Click the Instances tab.
2. Select the roles to restart.
3. Select Actions for Selected > Rolling Restart.
3. In the pop-up dialog box, select the options you want:
Restart only roles whose configurations are stale
Restart only roles that are running outdated software versions
Which role types to restart
4. If you select an HDFS, HBase, MapReduce, or YARN service, you can have their worker roles restarted in
batches. You can configure:
How many roles should be included in a batch - Cloudera Manager restarts the worker roles rack-by-rack
in alphabetical order, and within each rack, hosts are restarted in alphabetical order. If you are using the
default replication factor of 3, Hadoop tries to keep the replicas on at least 2 different racks. So if you
Cloudera Administration | 41

Managing CDH and Managed Services


have multiple racks, you can use a higher batch size than the default 1. But you should be aware that
using too high batch size also means that fewer worker roles are active at any time during the upgrade,
so it can cause temporary performance degradation. If you are using a single rack only, you should only
restart one worker node at a time to ensure data availability during upgrade.
How long should Cloudera Manager wait before starting the next batch.
The number of batch failures that will cause the entire rolling restart to fail (this is an advanced feature).
For example if you have a very large cluster you can use this option to allow failures because if you know
that your cluster will be functional even if some worker roles are down.
Note:
HDFS - If you do not have HDFS high availability configured, a warning appears reminding you
that the service will become unavailable during the restart while the NameNode is restarted.
Services that depend on that HDFS service will also be disrupted. It is recommended that you
restart the DataNodes one at a timeone host per batch, which is the default.
HBase
Administration operations such as any of the following should not be performed during the
rolling restart, to avoid leaving the cluster in an inconsistent state:

Split
Create, disable, enable, or drop table
Metadata changes
Create, clone, or restore a snapshot. Snapshots rely on the RegionServers being up;
otherwise the snapshot will fail.

To increase the speed of a rolling restart of the HBase service, set the Region Mover Threads
property to a higher value. This increases the number of regions that can be moved in
parallel, but places additional strain on the HMaster. In most cases, Region Mover Threads
should be set to 5 or lower.
Another option to increase the speed of a rolling restart of the HBase service is to set the
Skip Region Reload During Rolling Restart property to true. This setting can cause regions
to be moved around multiple times, which can degrade HBase client performance.
MapReduce - If you restart the JobTracker, all current jobs will fail.
YARN - If you restart ResourceManager and ResourceManager HA is enabled, current jobs
continue running: they do not restart or fail. ResourceManager HA is supported for CDH 5.2
and higher.
ZooKeeper and Flume - For both ZooKeeper and Flume, the option to restart roles in batches
is not available. They are always restarted one by one.
5. Click Confirm to start the rolling restart.
Performing a Cluster-Level Rolling Restart
You can perform a cluster-level rolling restart on demand from the Cloudera Manager Admin Console. A
cluster-level rolling restart is also performed as the last step in a rolling upgrade when the cluster is configured
with HDFS high availability enabled.
1. If you have not already done so, enable high availability. See HDFS High Availability on page 288 for instructions.
You do not need to enable automatic failover for rolling restart to work, though you can enable it if you wish.
Automatic failover does not affect the rolling restart operation.
2. For the cluster you want to restart select Actions > Rolling Restart.
3. In the pop-up dialog box, select the services you want to restart. Please review the caveats in the preceding
section for the services you elect to have restarted. The services that do not support rolling restart will simply
be restarted, and will be unavailable during their restart.

42 | Cloudera Administration

Managing CDH and Managed Services


4. If you select an HDFS, HBase, or MapReduce service, you can have their worker roles restarted in batches.
You can configure:
How many roles should be included in a batch - Cloudera Manager restarts the worker roles rack-by-rack
in alphabetical order, and within each rack, hosts are restarted in alphabetical order. If you are using the
default replication factor of 3, Hadoop tries to keep the replicas on at least 2 different racks. So if you
have multiple racks, you can use a higher batch size than the default 1. But you should be aware that
using too high batch size also means that fewer worker roles are active at any time during the upgrade,
so it can cause temporary performance degradation. If you are using a single rack only, you should only
restart one worker node at a time to ensure data availability during upgrade.
How long should Cloudera Manager wait before starting the next batch.
The number of batch failures that will cause the entire rolling restart to fail (this is an advanced feature).
For example if you have a very large cluster you can use this option to allow failures because if you know
that your cluster will be functional even if some worker roles are down.
5. Click Restart to start the rolling restart. While the restart is in progress, the Command Details page shows
the steps for stopping and restarting the services.

Aborting a Pending Command


Required Role:
Commands will time out if they are unable to complete after a period of time.
If necessary, you can abort a pending command. For example, this may become necessary because of a hardware
or network failure where a host running a role instance becomes disconnected from the Cloudera Manager
Server host. In this case, the Cloudera Manager Server will be unable to connect to the Cloudera Manager Agent
on that disconnected host to start or stop the role instance which will prevent the corresponding service from
starting or stopping. To work around this, you can abort the command to start or stop the role instance on the
disconnected host, and then you can start or stop the service again.
To abort any pending command:
You can click the indicator (
) with the blue badge, which shows the number of commands that are currently
running in your cluster (if any). This indicator is positioned just to the left of the Support link at the right hand
side of the navigation bar. Unlike the Commands tab for a role or service, this indicator includes all commands
running for all services or roles in the cluster. In the Running Commands window, click Abort to abort the pending
command. For more information, see Viewing Running and Recent Commands.
To abort a pending command for a service or role:
1. Navigate to the Service > Instances tab for the service where the role instance you want to stop is located.
For example, navigate to the HDFS Service > Instances tab if you want to abort a pending command for a
DataNode.
2. In the list of instances, click the link for role instance where the command is running (for example, the instance
that is located on the disconnected host).
3. Go to the Commands tab.
4. Find the command in the list of Running Commands and click Abort Command to abort the running command.

Deleting Services
Required Role:
1. Stop the service. For information on starting and stopping services, see Starting, Stopping, and Restarting
Services on page 39.
2.
On the Home > Status tab, click
to the right of the service name and select Delete.
3. Click Delete to confirm the deletion. Deleting a service does not clean up the associated client configurations
that have been deployed in the cluster or the user data stored in the cluster. For a given "alternatives path"

Cloudera Administration | 43

Managing CDH and Managed Services


(for example /etc/hadoop/conf) if there exist both "live" client configurations (ones that would be pushed
out with deploy client configurations for active services) and ones that have been "orphaned" client
configurations (the service they correspond to has been deleted), the orphaned ones will be removed from
the alternatives database. In other words, to trigger cleanup of client configurations associated with a deleted
service you must create a service to replace it. To remove user data, see Remove User Data.

Renaming a Service
Required Role:
A service is given a name upon installation, and that name is used as an identifier internally. However, Cloudera
Manager allows you to provide a display name for a service, and that name will appear in the Cloudera Manager
Admin Console instead of the original (internal) name.
1.

On the Home > Status tab, click


2. Type the new name.
3. Click Rename.

to the right of the service name and select Rename.

The original service name will still be used internally, and may appear or be required in certain circumstances,
such as in log messages or in the API.
The rename action is recorded as an Audit event.
When looking at Audit or Event search results for the renamed service, it is possible that these search results
might contain either only the original (internal) name, or both the display name and the original name.

Configuring Maximum File Descriptors


Required Role:
You can setting the maximum file descriptor parameter for all daemon roles. When not specified, the role uses
whatever value it inherits from supervisor. When specified, configures soft and hard limits to the configured
value.
1.
2.
3.
4.
5.
6.

Go to a service.
Click the Configuration tab.
In the Search box, type rlimit_fds.
Set the Maximum Process File Descriptors property for one or more roles.
Click Save Changes to commit the changes.
Restart the affected role instances.

Managing Roles
When Cloudera Manager configures a service, it configures hosts in your cluster with one or more functions
(called roles in Cloudera Manager) that are required for that service. The role determines which Hadoop daemons
run on a given host. For example, when Cloudera Manager configures an HDFS service instance it configures
one host to run the NameNode role, another host to run as the Secondary NameNode role, another host to run
the Balancer role, and some or all of the remaining hosts to run DataNode roles.
Configuration settings are organized in role groups. A role group includes a set of configuration properties for
a specific group, as well as a list of role instances associated with that role group. Cloudera Manager automatically
creates default role groups.
For role types that allow multiple instances on multiple hosts, such as DataNodes, TaskTrackers, RegionServers
(and many others), you can create multiple role groups to allow one set of role instances to use different
configuration settings than another set of instances of the same role type. In fact, upon initial cluster setup, if
you are installing on identical hosts with limited memory, Cloudera Manager will (typically) automatically create
two role groups for each worker role one group for the role instances on hosts with only other worker roles,
and a separate group for the instance running on the host that is also hosting master roles.
44 | Cloudera Administration

Managing CDH and Managed Services


The HDFS service is an example of this: Cloudera Manager typically creates one role group (DataNode Default
Group) for the DataNode role instances running on the worker hosts, and another group (HDFS-1-DATANODE-1)
for the DataNode instance running on the host that is also running the master roles such as the NameNode,
JobTracker, HBase Master and so on. Typically the configurations for those two classes of hosts will differ in
terms of settings such as memory for JVMs.
Cloudera Manager configuration screens offer two layout options: classic and new. The new layout is the default;
however, on each configuration page you can easily switch between layouts using the Switch to XXX layout link
at the top right of the page. For more information, see Configuration Overview on page 8.
Gateway Roles
A gateway is a special type of role whose sole purpose is to designate a host that should receive a client
configuration for a specific service, when the host does not have any roles running on it. Gateway roles enable
Cloudera Manager to install and manage client configurations on that host. There is no process associated with
a gateway role, and its status will always be Stopped. You can configure gateway roles for HBase, HDFS, Hive,
MapReduce, Solr, Spark, Sqoop 1 Client, and YARN.

Role Instances
Adding a Role Instance
Required Role:
After creating services, you can add role instances to the services. For example, after initial installation in which
you created the HDFS service, you can add a DataNode role instance to a host where one was not previously
running. Upon upgrading a cluster to a new version of CDH you might want to create a role instance for a role
added in the new version. For example, in CDH 5 Impala has the Impala Llama ApplicationMaster role, which
must be added after you upgrade a CDH 4 cluster to CDH 5.
1. Go to the service for which you want to add a role instance. For example, to add a DataNode role instance,
go to the HDFS service.
2. Click the Instances tab.
3. Click the Add Role Instances button.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Click Continue.
Cloudera Administration | 45

Managing CDH and Managed Services


6. In the Review Changes page, review the configuration changes to be applied. Confirm the settings entered
for file system paths. The file paths required vary based on the services to be installed. For example, you
might confirm the NameNode Data Directory and the DataNode Data Directory for HDFS. Click Continue. The
wizard finishes by performing any actions necessary to prepare the cluster for the new role instances. For
example, new DataNodes are added to the NameNode dfs_hosts_allow.txt file. The new role instance is
configured with the default role group for its role type, even if there are multiple role groups for the role type.
If you want to use a different role group, follow the instructions in Managing Role Groups on page 48 for
moving role instances to a different role group. The new role instances are not started automatically.
Starting, Stopping, and Restarting Role Instances
Required Role:
If the host for the role instance is currently decommissioned, you will not be able to start the role until the host
has been recommissioned.
1.
2.
3.
4.

Go to the service that contains the role instances to start, stop, or restart.
Click the Instances tab.
Check the checkboxes next to the role instances to start, stop, or restart (such as a DataNode instance).
Select Actions for Selected > Start, Stop, or Restart, and then click Start, Stop, or Restart again to start the
process. When you see a Finished status, the process has finished.

Also see Rolling Restart on page 41.


Decommissioning Role Instances
Required Role:
You can remove a role instance such as a DataNode from a cluster while the cluster is running by decommissioning
the role instance. When you decommission a role instance, Cloudera Manager performs a procedure so that you
can safely retire a host without losing data. Role decommissioning applies to HDFS DataNode, MapReduce
TaskTracker, YARN NodeManager, and HBase RegionServer roles.
You cannot decommission a DataNode or a host with a DataNode if the number of DataNodes equals the
replication factor (which by default is three) of any file stored in HDFS. For example, if the replication factor of
any file is three, and you have three DataNodes, you cannot decommission a DataNode or a host with a DataNode.
If you attempt to decommission a DataNode or a host with a DataNode in such situations, the DataNode will
be decommissioned, but the decommission process will not complete. You will have to abort the decommission
and recommission the DataNode.
A role will be decommissioned if its host is decommissioned. See Decommissioning and Recommissioning Hosts
on page 59 for more details.
To decommission role instances:
1. If you are decommissioning DataNodes, perform the steps in Tuning HDFS Prior to Decommissioning
DataNodes on page 60.
2. Click the service instance that contains the role instance you want to decommission.
3. Click the Instances tab.
4. Check the checkboxes next to the role instances to decommission.
5. Select Actions for Selected > Decommission, and then click Decommission again to start the process. A
Decommission Command pop-up displays that shows each step or decommission command as it is run. In
the Details area, click to see the subcommands that are run. Depending on the role, the steps may include
adding the host to an "exclusions list" and refreshing the NameNode, JobTracker, or NodeManager; stopping
the Balancer (if it is running); and moving data blocks or regions. Roles that do not have specific decommission
actions are stopped.
You can abort the decommission process by clicking the Abort button, but you must recommission and restart
the role.
46 | Cloudera Administration

Managing CDH and Managed Services


The Commission State facet in the Filters list displays

Decommissioning while decommissioning is in

progress, and
Decommissioned when the decommissioning process has finished. When the process is
complete, a
is added in front of Decommission Command.
Recommissioning Role Instances
Required Role:
1.
2.
3.
4.

Click the service that contains the role instance you want to recommission.
Click the Instances tab.
Check the checkboxes next to the decommissioned role instances to recommission.
Select Actions for Selected > Recommission, and then click Recommission to start the process. A Recommission
Command pop-up displays that shows each step or recommission command as it is run. When the process
is complete, a
is added in front of Recommission Command.
5. Restart the role instance.
Deleting Role Instances
Required Role:
1. Click the service instance that contains the role instance you want to delete. For example, if you want to
delete a DataNode role instance, click an HDFS service instance.
2. Click the Instances tab.
3. Check the checkboxes next to the role instances you want to delete.
4. If the role instance is running, select Actions for Selected > Stop and click Stop to confirm the action.
5. Select Actions for Selected > Delete. Click Delete to confirm the deletion.
Note: Deleting a role instance does not clean up the associated client configurations that have been
deployed in the cluster.
Configuring Roles to Use a Custom Garbage Collection Parameter
Every Java-based role in Cloudera Manager has a configuration setting called Java Configuration Options for
role where you can enter command line options. Commonly, garbage collection flags or extra debugging flags
would be passed here. To find the appropriate configuration setting, select the service you want to modify in
the Cloudera Manager Admin Console, then use the Search box to search for Java Configuration Options.
You can add configuration options for all instances of a given role by making this configuration change at the
service level. For example, to modify the setting for all DataNodes, select the HDFS service, then modify the Java
Configuration Options for DataNode setting.
To modify a configuration option for a given instance of a role, select the service, then select the particular role
instance (for example, a specific DataNode). The configuration settings you modify will apply to the selected role
instance only.
For detailed instructions see Modifying Configuration Properties Using Cloudera Manager on page 10.

Role Groups
Required Role:
A role group is a set of configuration properties for a role type, as well as a list of role instances associated with
that group. Cloudera Manager automatically creates a default role group named Role Type Default Group for
each role type.Each role instance can be associated with only a single role group.

Cloudera Administration | 47

Managing CDH and Managed Services


Role groups provide two types of properties: those that affect the configuration of the service itself and those
that affect monitoring of the service, if applicable (the Monitoring subcategory). (Not all services have monitoring
properties). For more information about monitoring properties see Configuring Monitoring Settings.
When you run the installation or upgrade wizard, Cloudera Manager configures the default role groups it adds,
and adds any other required role groups for a given role type. For example, a DataNode role on the same host
as the NameNode might require a different configuration than DataNode roles running on other hosts. Cloudera
Manager creates a separate role group for the DataNode role running on the NameNode host and uses the
default configuration for DataNode roles running on other hosts.
You can modify the settings of the default role group, or you can create new role groups and associate role
instances to whichever role group is most appropriate. This simplifies the management of role configurations
when one group of role instances may require different settings than another group of instances of the same
role typefor example, due to differences in the hardware the roles run on. You modify the configuration for
any of the service's role groups through the Configuration tab for the service. You can also override the settings
inherited from a role group for a role instance.
If there are multiple role groups for a role type, you can move role instances from one group to another. When
you move a role instance to a different group, it inherits the configuration settings for its new group.
Creating a Role Group
1.
2.
3.
4.
5.
6.

Go to a service status page.


Click the Instances or Configuration tab.
Click Role Groups.
Click Create new group....
Provide a name for the group.
Select the role type for the group. You can select role types that allow multiple instances and that exist for
the service you have selected.
7. In the Copy From field, select the source of the basic configuration information for the role group:
An existing role group of the appropriate type.
None.... The role group is set up with generic default values that are not the same as the values Cloudera
Manager sets in the default role group, as Cloudera Manager specifically sets the appropriate configuration
properties for the services and roles it installs. After you create the group you must edit the configuration
to set missing properties (for example the TaskTracker Local Data Directory List property, which is not
populated if you select None) and clear other validation warnings and errors.
Managing Role Groups
You can rename or delete existing role groups, and move roles of the same role type from one group to another.
1.
2.
3.
4.
5.

Go to a service status page.


Click the Instances or Configuration tab.
Click Role Groups.
Click the group you want to manage. Role instances assigned to the role group are listed.
Perform the appropriate procedure for the action:
Action

Procedure

Rename

1.

Delete

You cannot delete any of the default groups. The group must first be empty; if
you want to delete a group you've created, you must move any role instances to
a different role group.

Click the role group name, click


next to the name on the right and click
Rename.
2. Specify the new name and click Rename.

1. Click the role group name.

48 | Cloudera Administration

Managing CDH and Managed Services


Action

Procedure
2.

Move

Click
next to the role group name on the right, select Delete, and confirm
by clicking Delete. Deleting a role group removes it from host templates.

1. Select the role instance(s) to move.


2. Select Actions for Selected > Move To Different Role Group....
3. In the pop-up that appears, select the target role group and click Move.

Managing Hosts
Cloudera Manager provides a number of features that let you configure and manage the hosts in your clusters.
The Hosts screen has the following tabs:

The Status Tab


Viewing All Hosts
To display summary information about all the hosts managed by Cloudera Manager, click Hosts in the main
navigation bar. The All Hosts page displays with a list of all the hosts managed by Cloudera Manager.
The list of hosts shows the overall status of the Cloudera Manager-managed hosts in your cluster.
The information provided varies depending on which columns are selected. To change the columns, click the
Columns: n Selected drop-down and select the checkboxes next to the columns to display.
Clicking the to the left of the number of roles lists all the role instances running on that host. The balloon
annotation that appears when you move the cursor over a link indicates the service instance to which the
role belongs.
Filter the hosts by typing a property value in the Search box or selecting a value from the facets at the left
of the page. If the Configuring Agent Heartbeat and Health Status Options on page 433 are configured as
follows:
Send Agent heartbeat every x
Set health status to Concerning if the Agent heartbeats fail y
Set health status to Bad if the Agent heartbeats fail z
The value v for a host's Last Heartbeat facet is computed as follows:
v < x * y = Good
v >= x * y and <= x * z = Concerning
v >= x * z = Bad
Viewing the Hosts in a Cluster
Do one of the following:
Select Clusters > Cluster name > General > Hosts.
In the Home screen, click
in a full form cluster table.
The All Hosts page displays with a list of the hosts filtered by the cluster name.
Viewing Individual Hosts
You can view detailed information about an individual hostresources (CPU/memory/storage) used and
available, which processes it is running, details about the host agent, and much moreby clicking a host link
on the All Hosts page. See Viewing Host Details on page 50.

Cloudera Administration | 49

Managing CDH and Managed Services


The Configuration Tab
The Configuration tab lets you set properties related to parcels and to resource management, and also monitoring
properties for the hosts under management. The configuration settings you make here will affect all your
managed hosts. You can also configure properties for individual hosts from the Host Details page (see Viewing
Host Details on page 50) which will override the global properties set here).
To edit the Default configuration properties for hosts:
1. Click the Configuration tab.
For more information on making configuration changes, see Modifying Configuration Properties Using Cloudera
Manager on page 10.

The Roles and Disks Overview Tabs


Role Assignments
You can view the assignment of roles to hosts as follows:
1. Click the Roles tab.
2. Click a cluster name or All Clusters.
Disks Overview
Click the Disks Overview tab to display an overview of the status of all disks in the deployment. The statistics
exposed match or build on those in iostat, and are shown in a series of histograms that by default cover every
physical disk in the system.
Adjust the endpoints of the time line to see the statistics for different time periods. Specify a filter in the box
to limit the displayed data. For example, to see the disks for a single rack rack1, set the filter to:
logicalPartition = false and rackId = "rack1" and click Filter. Click a histogram to drill down and
identify outliers. Mouse over the graph and click

to display additional information about the chart.

The Templates Tab


The Templates tab lets you create and manage host templates, which provide a way to specify a set of role
configurations that should be applied to a host. This greatly simplifies the process of adding new hosts, because
it lets you specify the configuration for multiple roles on a host in a single step, and then (optionally) start all
those roles.

The Parcels Tab


In the Parcels tab you can download, distribute, and activate available parcels to your cluster. You can use parcels
to add new products to your cluster, or to upgrade products you already have installed.

Viewing Host Details


You can view detailed information about each host, including:
Name, IP address, rack ID
Health status of the host and last time the Cloudera Manager Agent sent a heartbeat to the Cloudera Manager
Server
Number of cores
System load averages for the past 1, 5, and 15 minutes
Memory usage
File system disks, their mount points, and usage
Health test results for the host
Charts showing a variety of metrics and health test results over time.
Role instances running on the host and their health
50 | Cloudera Administration

Managing CDH and Managed Services


CPU, memory, and disk resources used for each role instance
To view detailed host information:
1. Click the Hosts tab.
2. Click the name of one of the hosts. The Status page is displayed for the host you selected.
3. Click tabs to access specific categories of information. Each tab provides various categories of information
about the host, its services, components, and configuration.
From the status page you can view details about several categories of information.
Status
The Status page is displayed when a host is initially selected and provides summary information about the
status of the selected host. Use this page to gain a general understanding of work being done by the system,
the configuration, and health status.
If this host has been decommissioned or is in maintenance mode, you will see the following icon(s) (
the top bar of the page next to the status message.

) in

Details
This panel provides basic system configuration such as the host's IP address, rack, health status summary, and
disk and CPU resources. This information summarizes much of the detailed information provided in other panes
on this tab. To view details about the Host agent, click the Host Agent link in the Details section.
Health Tests
Cloudera Manager monitors a variety of metrics that are used to indicate whether a host is functioning as
expected. The Health Tests panel shows health test results in an expandable/collapsible list, typically with the
specific metrics that the test returned. (You can Expand All or Collapse All from the links at the upper right of
the Health Tests panel).
The color of the text (and the background color of the field) for a health test result indicates the status of the
results. The tests are sorted by their health status Good, Concerning, Bad, or Disabled. The list of entries
for good and disabled health tests are collapsed by default; however, Bad or Concerning results are shown
expanded.
The text of a health test also acts as a link to further information about the test. Clicking the text will pop
up a window with further information, such as the meaning of the test and its possible results, suggestions
for actions you can take or how to make configuration changes related to the test. The help text for a health
test also provides a link to the relevant monitoring configuration section for the service. See Configuring
Monitoring Settings for more information.
Health History
The Health History provides a record of state transitions of the health tests for the host.
Click the arrow symbol at the left to view the description of the health test state change.
Click the View link to open a new page that shows the state of the host at the time of the transition. In this
view some of the status settings are greyed out, as they reflect a time in the past, not the current status.
File Systems
The File systems panel provides information about disks, their mount points and usage. Use this information
to determine if additional disk space is required.
Roles
Use the Roles panel to see the role instances running on the selected host, as well as each instance's status
and health. Hosts are configured with one or more role instances, each of which corresponds to a service. The

Cloudera Administration | 51

Managing CDH and Managed Services


role indicates which daemon runs on the host. Some examples of roles include the NameNode, Secondary
NameNode, Balancer, JobTrackers, DataNodes, RegionServers and so on. Typically a host will run multiple roles
in support of the various services running in the cluster.
Clicking the role name takes you to the role instance's status page.
You can delete a role from the host from the Instances tab of the Service page for the parent service of the role.
You can add a role to a host in the same way. See Role Instances on page 45.
Charts
Charts are shown for each host instance in your cluster.
See Viewing Charts for Cluster, Service, Role, and Host Instances for detailed information on the charts that are
presented, and the ability to search and display metrics of your choice.
Processes
The Processes page provides information about each of the processes that are currently running on this host.
Use this page to access management web UIs, check process status, and access log information.
Note: The Processes page may display exited startup processes. Such processes are cleaned up
within a day.
The Processes tab includes a variety of categories of information.
Service - The name of the service. Clicking the service name takes you to the service status page. Using the
triangle to the right of the service name, you can directly access the tabs on the role page (such as the
Instances, Commands, Configuration, Audits, or Charts Library tabs).
Instance - The role instance on this host that is associated with the service. Clicking the role name takes
you to the role instance's status page. Using the triangle to the right of the role name, you can directly access
the tabs on the role page (such as the Processes, Commands, Configuration, Audits, or Charts Library tabs)
as well as the status page for the parent service of the role.
Name - The process name.
Links - Links to management interfaces for this role instance on this system. These is not available in all
cases.
Status - The current status for the process. Statuses include stopped, starting, running, and paused.
PID - The unique process identifier.
Uptime - The length of time this process has been running.
Full log file - A link to the full log (a file external to Cloudera Manager) for this host log entries for this host.
Stderr - A link to the stderr log (a file external to Cloudera Manager) for this host.
Stdout - A link to the stdout log (a file external to Cloudera Manager) for this host.
Resources
The Resources page provides information about the resources (CPU, memory, disk, and ports) used by every
service and role instance running on the selected host.
Each entry on this page lists:

The service name


The name of the particular instance of this service
A brief description of the resource
The amount of the resource being consumed or the settings for the resource

The resource information provided depends on the type of resource:


CPU - An approximate percentage of the CPU resource consumed.
Memory - The number of bytes consumed.
Disk - The disk location where this service stores information.

52 | Cloudera Administration

Managing CDH and Managed Services


Ports - The port number being used by the service to establish network connections.
Commands
The Commands page shows you running or recent commands for the host you are viewing. See Viewing Running
and Recent Commands for more information.
Configuration
Required Role:
The Configuration page for a host lets you set properties for the selected host. You can set properties in the
following categories:
Advanced - Advanced configuration properties. These include the Java Home Directory, which explicitly sets
the value of JAVA_HOME for all processes. This overrides the auto-detection logic that is normally used.
Monitoring - Monitoring properties for this host. The monitoring settings you make on this page will override
the global host monitoring settings you make on the Configuration tab of the Hosts page. You can configure
monitoring properties for:
health check thresholds
the amount of free space on the filesystem containing the Cloudera Manager Agent's log and process
directories
a variety of conditions related to memory usage and other properties
alerts for health check events
For some monitoring properties, you can set thresholds as either a percentage or an absolute value (in bytes).
Other - Other configuration properties.
Parcels - Configuration properties related to parcels. Includes the Parcel Director property, the directory that
parcels will be installed into on this host. If the parcel_dir variable is set in the Agent's config.ini file, it
will override this value.
Resource Management - Enables resource management using control groups (cgroups).
For more information, see the description for each or property or see Modifying Configuration Properties Using
Cloudera Manager on page 10.
Components
The Components page lists every component installed on this host. This may include components that have
been installed but have not been added as a service (such as YARN, Flume, or Impala).
This includes the following information:
Component - The name of the component.
Version - The version of CDH from which each component came.
Component Version - The detailed version number for each component.
Audits
The Audits page lets you filter for audit events related to this host. See Lifecycle and Security Auditing for more
information.
Charts Library
The Charts Library page for a host instance provides charts for all metrics kept for that host instance, organized
by category. Each category is collapsible/expandable. See Viewing Charts for Cluster, Service, Role, and Host
Instances for more information.

Using the Host Inspector


Required Role:

Cloudera Administration | 53

Managing CDH and Managed Services


You can use the host inspector to gather information about hosts that Cloudera Manager is currently managing.
You can review this information to better understand system status and troubleshoot any existing issues. For
example, you might use this information to investigate potential DNS misconfiguration.
The inspector runs tests to gather information for functional areas including:

Networking
System time
User and group configuration
HDFS settings
Component versions

Common cases in which this information is useful include:

Installing components
Upgrading components
Adding hosts to a cluster
Removing hosts from a cluster

Running the Host Inspector


1. Click the Hosts tab.
2. Click Host Inspector. Cloudera Manager begins several tasks to inspect the managed hosts.
3. After the inspection completes, click Download Result Data or Show Inspector Results to review the results.
The results of the inspection displays a list of all the validations and their results, and a summary of all the
components installed on your managed hosts.
If the validation process finds problems, the Validations section will indicate the problem. In some cases the
message may indicate actions you can take to resolve the problem. If an issue exists on multiple hosts, you may
be able to view the list of occurrences by clicking a small triangle that appears at the end of the message.
The Version Summary section shows all the components that are available from Cloudera, their versions (if
known) and the CDH distribution to which they belong (CDH 4 or CDH 5).
If you are running multiple clusters with both CDH 4 and CDH 5, the lists will be organized by distribution (CDH
4 or CDH 5). The hosts running that version are shown at the top of each list.
Viewing Past Host Inspector Results
You can view the results of a past host inspection by looking for the Host Inspector command using the Recent
Commands feature.
1.
2.
3.
4.
5.

Click the Running Commands indicator (


) just to the left of the Search box at the right hand side of the
navigation bar.
Click the Recent Commands button.
If the command is too far in the past, you can use the Time Range Selector to move the time range back to
cover the time period you want.
When you find the Host Inspector command, click its name to display its subcommands.
Click the Show Inspector Results button to view the report.

See Viewing Running and Recent Commands for more information about viewing past command activity.

Adding a Host to the Cluster


Required Role:
You can add one or more hosts to your cluster using the Add Hosts wizard, which installs the Oracle JDK, CDH,
and Cloudera Manager Agent software. After the software is installed and the Cloudera Manager Agent is started,
the Agent connects to the Cloudera Manager Server and you can use the Cloudera Manager Admin Console to
manage and monitor CDH on the new host.

54 | Cloudera Administration

Managing CDH and Managed Services


The Add Hosts wizard does not create roles on the new host; once you have successfully added the host(s) you
can either add roles, one service at a time, or apply a host template, which can define role configurations for
multiple roles.
Important:
All hosts in a single cluster must be running the same version of CDH.
When you add a new host, you must install the same version of CDH to enable the new host to
work with the other hosts in the cluster. The installation wizard lets you select the version of CDH
to install, and you can choose a custom repository to ensure that the version you install matches
the version on the other hosts.
If you are managing multiple clusters, select the version of CDH that matches the version in use
on the cluster where you plan to add the new host.
When you add a new host, the following occurs:
YARN topology.map is updated to include the new host
Any service that includes topology.map in its configurationFlume, Hive, Hue, Oozie, Solr,
Spark, Sqoop 2, YARNis marked stale
At a convenient point after adding the host you should restart the stale services to pick up the
new configuration.
Use one of the following methods to add a new host:
Using the Add Hosts Wizard to Add Hosts
You can use the Add Hosts wizard to install CDH, Impala, and the Cloudera Manager Agent on a host.
Disable TLS Encryption or Authentication
If you have enabled TLS encryption or authentication for the Cloudera Manager Agents, you must disable both
of them before starting the Add Hosts wizard. Otherwise, skip to the next step.
Important: This step leaves the existing hosts in an unmanageable state; they are still configured to
use TLS, and so can't communicate with the Cloudera Manager Server.
1. From the Administration tab, select Settings.
2. Select the Security category.
3. Disable all levels of TLS that are currently enabled by deselecting the following options: Use TLS Encryption
for Agents, and Use TLS Authentication of Agents to Server.
4. Click Save Changes to save the settings.
5. Restart the Cloudera Management Server to have these changes take effect.
Using the Add Hosts Wizard
1. Click the Hosts tab.
2. Click the Add New Hosts button.
3. Follow the instructions in the wizard to install the Oracle JDK and Cloudera Manager Agent packages and
start the Agent.
4. In the Specify hosts for your CDH Cluster installation page, you can search for new hosts to add under the
New Hosts tab. However, if you have hosts that are already known to Cloudera Manager but have no roles
assigned, (for example, a host that was previously in your cluster but was then removed) these will appear
under the Currently Managed Hosts tab.
5. You will have an opportunity to add (and start) role instances to your newly-added hosts using a host template.
a. You can select an existing host template, or create a new one.

Cloudera Administration | 55

Managing CDH and Managed Services


b. To create a new host template, click the + Create... button. This will open the Create New Host Template
pop-up. See Host Templates on page 57 for details on how you select the role groups that define the
roles that should run on a host. When you have created the template, it will appear in the list of host
templates from which you can choose.
c. Select the host template you want to use.
d. By default Cloudera Manager will automatically start the roles specified in the host template on your
newly added hosts. To prevent this, uncheck the option to start the newly-created roles.
6. When the wizard is finished, you can verify the Agent is connecting properly with the Cloudera Manager
Server by clicking the Hosts tab and checking the health status for the new host. If the Health Status is Good
and the value for the Last Heartbeat is recent, then the Agent is connecting properly with the Cloudera
Manager Server.
If you did not specify a host template during the Add Hosts wizard, then no roles will be present on your new
hosts until you add them. You can do this by adding individual roles under the Instances tab for a specific service,
or by using a host template. See Role Instances on page 45 for information about adding roles for a specific
service. See Host Templates on page 57 to create a host template that specifies a set of roles (from different
services) that should run on a host.
Enable TLS Encryption or Authentication
If you previously enabled TLS security on your cluster, you must re-enable the TLS options on the Administration
page and also configure TLS on each new host after using the Add Hosts wizard. Otherwise, you can ignore this
step. For instructions, see Configuring TLS Security for Cloudera Manager.
Enable TLS/SSL for CDH Components
If you have previously enabled TLS/SSL on your cluster, and you plan to start these roles on this new host, make
sure you install a new host certificate to be configured with the same path and naming convention as the rest
of your hosts. Because the new host and the roles configured on it are inheriting their configuration from the
previous host, ensure that the keystore or truststore passwords and locations are the same on the new host.
For instructions on configuring TLS/SSL, see Configuring TLS/SSL Encryption for CDH Services.
Enable Kerberos
If you have previously enabled Kerberos on your cluster:
Install the packages required to kinit on the new host. For the list of packages required for each OS, see
Kerberos Prerequisites.
If you have set up Cloudera Manager to manage krb5.conf, it will automatically deploy the file on the new
host.
If Cloudera Manager does not manage krb5.conf, you must manually update the file at /etc/krb5.conf.
Adding a Host by Installing the Packages Using Your Own Method
If you used a different mechanism to install the Oracle JDK, CDH, Cloudera Manager Agent packages, you can
use that same mechanism to install the Oracle JDK, CDH, Cloudera Manager Agent packages and then start the
Cloudera Manager Agent.
1. Install the Oracle JDK, CDH, and Cloudera Manager Agent packages using your own method. For instructions
on installing these packages, see Installation Path B - Manual Installation Using Cloudera Manager Packages.
2. After installation is complete, start the Cloudera Manager Agent. For instructions, see Starting, Stopping, and
Restarting Cloudera Manager Agents on page 431.
3. After the Agent is started, you can verify the Agent is connecting properly with the Cloudera Manager Server
by clicking the Hosts tab and checking the health status for the new host. If the Health Status is Good and
the value for the Last Heartbeat is recent, then the Agent is connecting properly with the Cloudera Manager
Server.
4. If you have enabled TLS security on your cluster, you must enable and configure TLS on each new host.
Otherwise, ignore this step.
56 | Cloudera Administration

Managing CDH and Managed Services


a. Enable and configure TLS on each new host by specifying 1 for the use_tls property in the
/etc/cloudera-scm-agent/config.ini configuration file.
b. Configure the same level(s) of TLS security on the new hosts by following the instructions in Configuring
TLS Security for Cloudera Manager.
5. If you have previously enabled TLS/SSL on your cluster, and you plan to start these roles on this new host,
make sure you install a new host certificate to be configured with the same path and naming convention as
the rest of your hosts. Because the new host and the roles configured on it are inheriting their configuration
from the previous host, ensure that the keystore or truststore passwords and locations are the same on the
new host. For instructions on configuring TLS/SSL, see Configuring TLS/SSL Encryption for CDH Services.
6. If you have previously enabled Kerberos on your cluster:
Install the packages required to kinit on the new host. For the list of packages required for each OS, see
Kerberos Prerequisites.
If you have set up Cloudera Manager to manage krb5.conf, it will automatically deploy the file on the
new host.
If Cloudera Manager does not manage krb5.conf, you must manually update the file at /etc/krb5.conf.

Specifying Racks for Hosts


Required Role:
To get maximum performance, it is important to configure CDH so that it knows the topology of your network.
Network locations such as hosts and racks are represented in a tree, which reflects the network distance
between locations. HDFS will use the network location to be able to place block replicas more intelligently to
trade off performance and resilience. When placing jobs on hosts, CDH will prefer within-rack transfers (where
there is more bandwidth available) to off-rack transfers; the MapReduce and YARN schedulers use network
location to determine where the closest replica is as input to a map task. These computations are performed
with the assistance of rack awareness scripts.
Cloudera Manager includes internal rack awareness scripts, but you must specify the racks where the hosts in
your cluster are located. If your cluster contains more than 10 hosts, Cloudera recommends that you specify the
rack for each host. HDFS, MapReduce, and YARN will automatically use the racks you specify.
Cloudera Manager supports nested rack specifications. For example, you could specify the rack /rack3, or
/group5/rack3 to indicate the third rack in the fifth group. All hosts in a cluster must have the same number
of path components in their rack specifications.
To specify racks for hosts:
1.
2.
3.
4.

Click the Hosts tab.


Check the checkboxes next to the host(s) for a particular rack, such as all hosts for /rack123.
Click Actions for Selected (n) > Assign Rack, where n is the number of selected hosts.
Enter a rack name or ID that starts with a slash /, such as /rack123 or /aisle1/rack123, and then click
Confirm.
5. Optionally restart affected services. Rack assignments are not automatically updated for running services.

Host Templates
Required Role:
Host templates let you designate a set of role groups that can be applied in a single operation to a host or a set
of hosts. This significantly simplifies the process of configuring new hosts when you need to expand your cluster.
Host templates are supported for both CDH 4 and CDH 5 cluster hosts.
Important: A host template can only be applied on a host with a version of CDH that matches the
CDH version running on the cluster to which the host template belongs.

Cloudera Administration | 57

Managing CDH and Managed Services


You can create and manage host templates under the Templates tab from the Hosts page.
1. Click the Hosts tab on the main Cloudera Manager navigation bar.
2. Click the Templates tab on the Hosts page.
Templates are not required; Cloudera Manager assigns roles and role groups to the hosts of your cluster when
you perform the initial cluster installation. However, if you want to add new hosts to your cluster, a host template
can make this much easier.
If there are existing host templates, they are listed on the page, along with links to each role group included in
the template.
If you are managing multiple clusters, you must create separate host templates for each cluster, as the templates
specify role configurations specific to the roles in a single cluster. Existing host templates are listed under the
cluster to which they apply.
You can click a role group name to be taken to the Edit configuration page for that role group, where you can
modify the role group settings.
From the Actions menu associated with the template you can edit the template, clone it, or delete it.
Creating a Host Template
1. From the Templates tab, click Click here
2. In the Create New Host Template pop-up window that appears:
Type a name for the template.
For each role, select the appropriate role group. There may be multiple role groups for a given role type
you want to select the one with the configuration that meets your needs.
3. Click Create to create the host template.
Editing a Host Template
1. From the Hosts tab, click the Templates tab.
2. Pull down the Actions menu for the template you want to modify, and click Edit. This put you into the Edit
Host Template pop-up window. This works exactly like the Create New Host Template window you can
modify they template name or any of the role group selections.
3. Click OK when you have finished.
Applying a Host Template to a Host
You can use a host template to apply configurations for multiple roles in a single operation.
You can apply a template to a host that has no roles on it, or that has roles from the same services as those
included in the host template. New roles specified in the template that do not already exist on the host will be
added. A role on the host that is already a member of the role group specified in the template will be left
unchanged. If a role on the host matches a role in the template, but is a member of a different role group, it will
be moved to the role group specified by the template.
For example, suppose you have two role groups for a DataNode (DataNode Default Group and DataNode (1)).
The host has a DataNode role that belongs to DataNode Default Group. If you apply a host template that specifies
the DataNode (1) group, the role on the host will be moved from DataNode Default Group to DataNode (1).
However, if you have two instances of a service, such as MapReduce (for example, mr1 and mr2) and the host
has a TaskTracker role from service mr2, you cannot apply a TaskTracker role from service mr1.
A host may have no roles on it if you have just added the host to your cluster, or if you decommissioned a
managed host and removed its existing roles.
Also, the host must have the same version of CDH installed as is running on the cluster whose host templates
you are applying.
If a host belongs to a different cluster than the one for which you created the host template, you can apply the
host template if the "foreign" host either has no roles on it, or has only management roles on it. When you apply

58 | Cloudera Administration

Managing CDH and Managed Services


the host template, the host will then become a member of the cluster whose host template you applied. The
following instructions assume you have already created the appropriate host template.
1.
2.
3.
4.
5.

Go to the Hosts page, Status tab.


Select the host(s) to which you want to apply your host template.
From the Actions for Selected menu, select Apply Host Template.
In the pop-up window that appears, select the host template you want to apply.
Optionally you can have Cloudera Manager start the roles created per the host template check the box to
enable this.
6. Click Confirm to initiate the action.

Decommissioning and Recommissioning Hosts


Decommissioning a host decommissions and stops all roles on the host without requiring you to individually
decommission the roles on each service. Decommissioning applies to only to HDFS DataNode, MapReduce
TaskTracker, YARN NodeManager, and HBase RegionServer roles. If the host has other roles running on it, those
roles are stopped.
After all roles on the host have been decommissioned and stopped, the host can be removed from service. You
can decommission multiple hosts in parallel.
Decommissioning Hosts
Required Role:
You cannot decommission a DataNode or a host with a DataNode if the number of DataNodes equals the
replication factor (which by default is three) of any file stored in HDFS. For example, if the replication factor of
any file is three, and you have three DataNodes, you cannot decommission a DataNode or a host with a DataNode.
If you attempt to decommission a DataNode or a host with a DataNode in such situations, the DataNode will
be decommissioned, but the decommission process will not complete. You will have to abort the decommission
and recommission the DataNode.
To decommission hosts:
1. If the host has a DataNode, perform the steps in Tuning HDFS Prior to Decommissioning DataNodes on page
60.
2. Click the Hosts tab.
3. Select the checkboxes next to one or more hosts.
4. Select Actions for Selected > Hosts Decommission.
A confirmation pop-up informs you of the roles that will be decommissioned or stopped on the hosts you
have selected.
5. Click Confirm. A Decommission Command pop-up displays that shows each step or decommission command
as it is run, service by service. In the Details area, click to see the subcommands that are run for
decommissioning a given service. Depending on the service, the steps may include adding the host to an
"exclusions list" and refreshing the NameNode, JobTracker, or NodeManager; stopping the Balancer (if it is
running); and moving data blocks or regions. Roles that do not have specific decommission actions are
stopped.
You can abort the decommission process by clicking the Abort button, but you must recommission and restart
each role that has been decommissioned.
The Commission State facet in the Filters lists displays

Decommissioning while decommissioning is in

progress, and
Decommissioned when the decommissioning process has finished. When the process is
complete, a
is added in front of Decommission Command.
You cannot start roles on a decommissioned host.
Cloudera Administration | 59

Managing CDH and Managed Services


Note: When a DataNode is decommissioned, the data blocks are not removed from the storage
directories. You must delete the data manually.
Tuning HDFS Prior to Decommissioning DataNodes
Required Role:
When a DataNode is decommissioned, the NameNode ensures that every block from the DataNode will still be
available across the cluster as dictated by the replication factor. This procedure involves copying blocks from
the DataNode in small batches. If a DataNode has thousands of blocks, decommissioning can take several hours.
Before decommissioning hosts with DataNodes, you should first tune HDFS:
1. Raise the heap size of the DataNodes. DataNodes should be configured with at least 4 GB heap size to allow
for the increase in iterations and max streams.
a.
b.
c.
d.
e.

Go to the HDFS service page.


Click the Configuration tab.
Select Scope > DataNode.
Select Category > Resource Management.
Set the Java Heap Size of DataNode in Bytes property as recommended.
If more than one role group applies to this configuration, edit the value for the appropriate role group.
See Modifying Configuration Properties Using Cloudera Manager on page 10.

f. Click Save Changes to commit the changes.


2. Set the DataNode balancing bandwidth:
a. Select Scope > DataNode.
b. Expand the Category > Performance category.
c. Configure the DataNode Balancing Bandwidth property to the bandwidth you have on your disks and
network.
d. Click Save Changes to commit the changes.
3. Increase the replication work multiplier per iteration to a larger number (the default is 2, however 10 is
recommended):
a. Select Scope > NameNode.
b. Expand the Category > Advanced category.
c. Configure the Replication Work Multiplier Per Iteration property to a value such as 10.
If more than one role group applies to this configuration, edit the value for the appropriate role group.
See Modifying Configuration Properties Using Cloudera Manager on page 10.
d. Click Save Changes to commit the changes.
4. Increase the replication maximum threads and maximum replication thread hard limits:
a. Select Scope > NameNode.
b. Expand the Category > Advanced category.
c. Configure the Maximum number of replication threads on a DataNode and Hard limit on the number of
replication threads on a DataNode properties to 50 and 100 respectively.
If more than one role group applies to this configuration, edit the value for the appropriate role group.
See Modifying Configuration Properties Using Cloudera Manager on page 10.
d. Click Save Changes to commit the changes.
5. Restart the HDFS service.

60 | Cloudera Administration

Managing CDH and Managed Services


Tuning HBase Prior to Decommissioning DataNodes
Required Role:
To increase the speed of a rolling restart of the HBase service, set the Region Mover Threads property to a higher
value. This increases the number of regions that can be moved in parallel, but places additional strain on the
HMaster. In most cases, Region Mover Threads should be set to 5 or lower.
Recommissioning Hosts
Required Role:
Only hosts that are decommissioned using Cloudera Manager can be recommissioned.
1. Click the Hosts tab.
2. Select one or more hosts to recommission.
3. Select Actions for Selected > Recommission and Confirm. A Recommission Command pop-up displays that
shows each step or recommission command as it is run. When the process is complete, a is added in front
of Recommission Command. The host and roles are marked as commissioned, but the roles themselves are
not restarted.
Restarting All The Roles on a Host
Required Role:
1. Click the Hosts tab.
2. Select one or more hosts on which to start all roles.
3. Select Actions for Selected > Start Roles on Hosts.

Deleting Hosts
Required Role:
You can remove a host from a cluster in two ways:
Delete the host entirely from Cloudera Manager.
Remove a host from a cluster, but leave it available to other clusters managed by Cloudera Manager.
Both methods decommission the hosts, delete roles, and remove managed service software, but preserve data
directories.
Deleting a Host from Cloudera Manager
1.
2.
3.
4.

In the Cloudera Manager Admin Console, click the Hosts tab.


Select the hosts to delete.
Select Actions for Selected > Decommission.
Stop the Agent on the host. For instructions, see Starting, Stopping, and Restarting Cloudera Manager Agents
on page 431.
5. In the Cloudera Manager Admin Console, click the Hosts tab.
6. Reselect the hosts you selected in step 2.
7. Select Actions for Selected > Delete.
Removing a Host From a Cluster
This procedure leaves the host managed by Cloudera Manager and preserves the Cloudera Management Service
roles (such as the Events Server, Activity Monitor, and so on).
1. In the Cloudera Manager Admin Console, click the Hosts tab.
Cloudera Administration | 61

Managing CDH and Managed Services


2. Select the hosts to delete.
3. Select Actions for Selected > Remove From Cluster. The Remove Hosts From Cluster dialog displays.
4. Leave the selections to decommission roles and skip removing the Cloudera Management Service roles. Click
Confirm to proceed with removing the selected hosts.

Maintenance Mode
Required Role:
Maintenance mode allows you to suppress alerts for a host, service, role, or an entire cluster. This can be useful
when you need to take actions in your cluster (make configuration changes and restart various elements) and
do not want to see the alerts that will be generated due to those actions.
Putting an entity into maintenance mode does not prevent events from being logged; it only suppresses the
alerts that those events would otherwise generate. You can see a history of all the events that were recorded
for entities during the period that those entities were in maintenance mode.
Explicit and Effective Maintenance Mode
When you enter maintenance mode on an entity (cluster, service, or host) that has subordinate entities (for
example, the roles for a service) the subordinate entities are also put into maintenance mode. These are considered
to be in effective maintenance mode, as they have inherited the setting from the higher-level entity.
For example:
If you set the HBase service into maintenance mode, then its roles (HBase Master and all RegionServers) are
put into effective maintenance mode.
If you set a host into maintenance mode, then any roles running on that host are put into effective
maintenance mode.
Entities that have been explicitly put into maintenance mode show the icon

. Entities that have entered

effective maintenance mode as a result of inheritance from a higher-level entity show the icon

When an entity (role, host or service) is in effective maintenance mode, it can only be removed from maintenance
mode when the higher-level entity exits maintenance mode. For example, if you put a service into maintenance
mode, then the roles associated with that service will be entered into effective maintenance mode, and will
remain in effective maintenance mode until the service exits maintenance mode. You cannot remove them from
maintenance mode individually.
On the other hand, an entity that is in effective maintenance mode can be put into explicit maintenance mode.
In this case, the entity will remain in maintenance mode even when the higher-level entity exits maintenance
mode. For example, suppose you put a host into maintenance mode, (which puts all the roles on that host into
effective maintenance mode). You then select one of the roles on that host and put it explicitly into maintenance
mode. When you have the host exit maintenance mode, that one role will remain in maintenance mode. You
will need to select it individually and specifically have it exit maintenance mode.

Viewing Maintenance Mode Status


You can view the status of Maintenance Mode in your cluster by clicking
selecting View Maintenance Mode Status.

to the right of the cluster name and

Entering Maintenance Mode


You can enable maintenance mode for a cluster, service, role, or host.
Putting a Cluster into Maintenance Mode
1.

Click

to the right of the cluster name and select Enter Maintenance Mode.

62 | Cloudera Administration

Managing CDH and Managed Services


2. Confirm that you want to do this.
The cluster is put into explicit maintenance mode, as indicated by the
cluster are entered into effective maintenance mode, as indicated by the

icon. All services and roles in the


icon.

Putting a Service into Maintenance Mode


1.

Click
to the right of the service name and select Enter Maintenance Mode.
2. Confirm that you want to do this.
The service is put into explicit maintenance mode, as indicated by the
entered into effective maintenance mode, as indicated by the

icon. All roles for the service are

icon.

Putting Roles into Maintenance Mode


1.
2.
3.
4.
5.

Go to the service page that includes the role.


Go to the Instances tab.
Select the role(s) you want to put into maintenance mode.
From the Actions for Selected menu, select Enter Maintenance Mode.
Confirm that you want to do this.

The roles will be put in explicit maintenance mode. If the roles were already in effective maintenance mode
(because its service or host was put into maintenance mode) the roles will now be in explicit maintenance mode.
This means that they will not exit maintenance mode automatically if their host or service exits maintenance
mode; they must be explicitly removed from maintenance mode.
Putting a Host into Maintenance Mode
1.
2.
3.
4.

Go to the Hosts page.


Select the host(s) you want to put into maintenance mode.
From the Actions for Selected menu, select Enter Maintenance Mode.
Confirm that you want to do this.

The confirmation pop-up lists the role instances that will be put into effective maintenance mode when the
host goes into maintenance mode.

Exiting Maintenance Mode


When you exit maintenance mode, the maintenance mode icons are removed and alert notification resumes.
Exiting a Cluster from Maintenance Mode
1.

Click
to the right of the cluster name and select Exit Maintenance Mode.
2. Confirm that you want to do this.
Exiting a Service from Maintenance Mode
1.

Click
to the right of the service name and select Exit Maintenance Mode.
2. Confirm that you want to do this.
Exiting Roles from Maintenance Mode
1. Go to the services page that includes the role.
2. Go to the Instances tab.
3. Select the role(s) you want to exit from maintenance mode.
Cloudera Administration | 63

Managing CDH and Managed Services


4. From the Actions for Selected menu, select Exit Maintenance Mode.
5. Confirm that you want to do this.
Exiting a Host from Maintenance Mode
1.
2.
3.
4.

Go to the Hosts page.


Select the host(s) you want to put into maintenance mode.
From the Actions for Selected menu, select Exit Maintenance Mode.
Confirm that you want to do this.

The confirmation pop-up lists the role instances that will be removed from effective maintenance mode when
the host exits maintenance mode.

Managing CDH from the Command Line


The following sections provide instructions and information on managing core Hadoop.
For installation and upgrade instructions, see the Cloudera Installation and Upgrade guide, which also contains
initial deployment and configuration instructions for core Hadoop and the CDH components, including:
Cluster configuration and maintenance:

Ports Used by Components of CDH 5


Configuring Network Names
Deploying CDH 5 on a Cluster
Starting CDH Services on page 65
Stopping CDH Services Using the Command Line on page 70

Avro Usage
Flume configuration
HBase configuration:
Configuration Settings for HBase
HBase Replication on page 405
HBase Snapshots
HCatalog configuration
Impala configuration
Hive configuration:
Configuring the Hive Metastore
Configuring HiveServer2
Configuring the Metastore to use HDFS High Availability
HttpFS configuration
Hue: Configuring CDH Components for Hue
Oozie configuration:
Configuring Oozie
Configuring Oozie Failover (hot/cold)
Parquet: Using the Parquet File Format with Impala, Hive, Pig, HBase, and MapReduce
Snappy:

Using Snappy for MapReduce Compression


Using Snappy for Pig Compression
Using Snappy for Hive Compression
Using Snappy Compression in Sqoop 1 and Sqoop 2 Imports

64 | Cloudera Administration

Managing CDH and Managed Services


Using Snappy Compression with HBase
Spark configuration:
Managing Spark Standalone Using the Command Line on page 222
Running Spark Applications on page 234
Running a Crunch Application with Spark on page 242
Sqoop configuration:
Setting HADOOP_MAPRED_HOME for Sqoop
Configuring Sqoop 2
ZooKeeper: Maintaining a ZooKeeper Server

Starting CDH Services


You need to start and stop services in the right order to make sure everything starts or stops cleanly.
Note: The Oracle JDK is required for all Hadoop components.
START services in this order:
Order

Service

Comments

For instructions and more


information

ZooKeeper

Cloudera recommends
starting ZooKeeper before
starting HDFS; this is a
requirement in a
high-availability (HA)
deployment. In any case,
always start ZooKeeper
before HBase.

Installing the ZooKeeper


Server Package and
Starting ZooKeeper on a
Single Server; Installing
ZooKeeper in a Production
Environment; Deploying
HDFS High Availability on
page 301; Configuring High
Availability for the
JobTracker (MRv1)

HDFS

Start HDFS before all other Deploying HDFS on a


services except ZooKeeper. Cluster; HDFS High
If you are using HA, see the Availability on page 288
CDH 5 High Availability
Guide for instructions.

HttpFS

4a

MRv1

Start MapReduce before Deploying MapReduce v1


Hive or Oozie. Do not start (MRv1) on a Cluster;
MRv1 if YARN is running. Configuring High
Availability for the
JobTracker (MRv1)

4b

YARN

Start YARN before Hive or Deploying MapReduce v2


Oozie. Do not start YARN (YARN) on a Cluster
if MRv1 is running.

HBase

HttpFS Installation

Starting and Stopping


HBase on page 82;
Deploying HBase in a
Distributed Cluster

Cloudera Administration | 65

Managing CDH and Managed Services


Order

Service

Comments

For instructions and more


information

Hive

Start the Hive metastore Installing Hive


before starting
HiveServer2 and the Hive
console.

Oozie

Starting the Oozie Server

Flume 1.x

Running Flume

Sqoop

Sqoop Installation and


Sqoop 2 Installation

10

Hue

Hue Installation

Configuring init to Start Hadoop System Services


Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
init(8) starts some daemons when the system is booted. Depending on the distribution, init executes scripts
from either the /etc/init.d directory or the /etc/rc2.d directory. The CDH packages link the files in init.d
and rc2.d so that modifying one set of files automatically updates the other.

To start system services at boot time and on restarts, enable their init scripts on the systems on which the
services will run, using the appropriate tool:
chkconfig is included in the RHEL and CentOS distributions. Debian and Ubuntu users can install the
chkconfig package.
update-rc.d is included in the Debian and Ubuntu distributions.
Configuring init to Start Core Hadoop System Services in an MRv1 Cluster
Important:
Cloudera does not support running MRv1 and YARN daemons on the same nodes at the same time;
it will degrade performance and may result in an unstable cluster deployment.
The chkconfig commands to use are:
$ sudo chkconfig hadoop-hdfs-namenode on

The update-rc.d commands to use on Ubuntu and Debian systems are:

66 | Cloudera Administration

Managing CDH and Managed Services


Where
On the NameNode

On the JobTracker

On the Secondary NameNode (if used)

On each TaskTracker

On each DataNode

Command
$ sudo update-rc.d hadoop-hdfs-namenode
defaults

$ sudo update-rc.d
hadoop-0.20-mapreduce-jobtracker defaults

$ sudo update-rc.d
hadoop-hdfs-secondarynamenode defaults

$ sudo update-rc.d
hadoop-0.20-mapreduce-tasktracker defaults

$ sudo update-rc.d hadoop-hdfs-datanode


defaults

Configuring init to Start Core Hadoop System Services in a YARN Cluster


Important:
Do not run MRv1 and YARN on the same set of nodes at the same time. This is not recommended; it
degrades your performance and may result in an unstable MapReduce cluster deployment.
The chkconfig commands to use are:
Where
On the NameNode
On the ResourceManager

On the Secondary NameNode (if used)

On each NodeManager

On each DataNode
On the MapReduce JobHistory node

Command
$ sudo chkconfig hadoop-hdfs-namenode on

$ sudo chkconfig
hadoop-yarn-resourcemanager on

$ sudo chkconfig
hadoop-hdfs-secondarynamenode on

$ sudo chkconfig hadoop-yarn-nodemanager


on

$ sudo chkconfig hadoop-hdfs-datanode on

$ sudo chkconfig
hadoop-mapreduce-historyserver on

The update-rc.d commands to use on Ubuntu and Debian systems are:

Cloudera Administration | 67

Managing CDH and Managed Services


Where

Command

On the NameNode

$ sudo update-rc.d hadoop-hdfs-namenode


defaults

On the ResourceManager

$ sudo update-rc.d
hadoop-yarn-resourcemanager defaults

On the Secondary NameNode (if used)

On each NodeManager

$ sudo update-rc.d
hadoop-hdfs-secondarynamenode defaults

$ sudo update-rc.d hadoop-yarn-nodemanager


defaults

On each DataNode

$ sudo update-rc.d hadoop-hdfs-datanode


defaults

On the MapReduce JobHistory node

$ sudo update-rc.d
hadoop-mapreduce-historyserver defaults

Configuring init to Start Non-core Hadoop System Services


Non-core Hadoop daemons can also be configured to start at init time using the chkconfig or update-rc.d
command.
The chkconfig commands are:
Component

Server

Hue

Hue server

$ sudo chkconfig hue on

Oozie

Oozie server

$ sudo chkconfig oozie on

HBase

HBase master

On each HBase RegionServer

Hive Metastore

Hive Metastore server

HiveServer2

HiveServer2

68 | Cloudera Administration

Command

$ sudo chkconfig
hbase-master on

$ sudo chkconfig
hbase-regionserver on

$ sudo chkconfig
hive-metastore on
$ sudo chkconfig hive-server2
on

Managing CDH and Managed Services


Component

Server

Zookeeper

Zookeeper server

HttpFS

HttpFS server

Command
$ sudo chkconfig
zookeeper-server on

$ sudo chkconfig
hadoop-httpfs on

The update-rc.d commands to use on Ubuntu and Debian systems are:


Component

Server

Hue

Hue server

Oozie

Oozie server

HBase

HBase master

HBase RegionServer

Hive Metastore

Hive Metastore server

HiveServer2

HiveServer2

Zookeeper

Zookeeper server

HttpFS

HttpFS server

Command
$ sudo update-rc.d hue
defaults

$ sudo update-rc.d oozie


defaults

$ sudo update-rc.d
hbase-master defaults

$ sudo update-rc.d
hbase-regionserver defaults

$ sudo update-rc.d
hive-metastore defaults
$ sudo update-rc.d
hive-server2 defaults
$ sudo update-rc.d
zookeeper-server defaults

$ sudo update-rc.d
hadoop-httpfs defaults

Starting and Stopping HBase Using the Command Line


When starting and stopping CDH services, order is important. See Starting CDH Services on page 65 and Stopping
CDH Services Using the Command Line on page 70 for details. If you use Cloudera Manager, follow these
instructions instead.
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Starting HBase
When starting HBase, it is important to start the HMaster, followed by the RegionServers, then the Thrift server.

Cloudera Administration | 69

Managing CDH and Managed Services


1. To start a HBase cluster using the command line, start the HBase Master by using the sudo hbase-master
start command on RHEL or SuSE, or the sudo hadoop-hbase-regionserver start command on Ubuntu
or Debian. The HMaster starts the RegionServers automatically.
2. To start a RegionServer manually, use the sudo hbase-regionserver start command on RHEL or SuSE,
or the sudo hadoop-hbase-regionserver start command on Ubuntu or Debian.
3. To start the Thrift server, use the hbase-thrift start on RHEL or SuSE, or the hadoop-hbase-thrift
start on Ubuntu or Debian.
Stopping HBase
When stopping HBase, it is important to stop the Thrift server, followed by each RegionServer, followed by any
backup HMasters, and finally the main HMaster.
1. Shut down the Thrift server by using the hbase-thrift stop command on the Thrift server host. sudo
service hbase-thrift stop

2. Shut down each RegionServer by using the hadoop-hbase-regionserver stop command on the
RegionServer host.
sudo service hadoop-hbase-regionserver stop

3. Shut down backup HMasters, followed by the main HMaster, by using the hbase-master stop command.
sudo service hbase-master stop

Stopping CDH Services Using the Command Line


Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Run the following command on every host in the cluster to shut down all Hadoop Common system services that
are started by init in the cluster:
$ for x in `cd /etc/init.d ; ls hadoop-*` ; do sudo service $x stop ; done

To verify that no Hadoop processes are running, issue the following command on each host::
# ps -aef | grep java

You could also use ps -fu hdfs and ps -fu mapred to confirm that no processes are running as the hdfs or
mapred user, but additional user names are created by the other components in the ecosystem; checking for
the "java" string in the ps output is an easy way to identify any processes that may still be running.
To stop system services individually, use the instructions in the table below.
STOP system services in this order:
Order

Service

Comments

Instructions

Hue

Run the following on the Hue Server machine to stop


Hue sudo service hue stop

Sqoop 1

Run the following on all nodes where it is running: sudo


service sqoop-metastore stop

70 | Cloudera Administration

Managing CDH and Managed Services


Order

Service

Sqoop 2

Comments

Instructions
Run the following on all nodes where it is running: $
sudo /sbin/service sqoop2-server stop

Flume 0.9

Stop the Flume Node processes on each node where


they are running:
sudo service flume-node stop Stop the Flume
Master sudo service flume-master stop

Flume 1.x

There is no Flume master Stop the Flume Node processes on each node where
they are running:
sudo service flume-ng-agent stop

Oozie

Hive

sudo service oozie stop

To stop Hive, exit the Hive console and make sure no


Hive scripts are running.
Shut down HiveServer2: sudo service hiveserver2
stop

Shut down the Hive metastore daemon on each client:


sudo service hive-metastore stop

If the metastore is running from the command line,


use Ctrl-c to shut it down.
7

HBase

Stop the Thrift server and See Starting and Stopping HBase on page 82.
clients, then shut down the
cluster.

8a

MapReduce v1

Stop Hive and Oozie before To stop MapReduce, stop the JobTracker service, and
stopping MapReduce.
stop the Task Tracker on all nodes where it is running.
Use the following commands:
sudo service
hadoop-0.20-mapreduce-jobtracker stop
sudo service
hadoop-0.20-mapreduce-tasktracker stop

8b

YARN

Stop Hive and Oozie before To stop YARN, stop the MapReduce JobHistory service,
stopping YARN.
ResourceManager service, and NodeManager on all
nodes where they are running. Use the following
commands:
sudo service hadoop-mapreduce-historyserver
stop
sudo service hadoop-yarn-resourcemanager
stop
sudo service hadoop-yarn-nodemanager stop

HttpFS

10

HDFS

sudo service hadoop-httpfs stop

To stop HDFS: On the NameNode: sudo service


hadoop-hdfs-namenode stop

Cloudera Administration | 71

Managing CDH and Managed Services


Order

Service

Comments

Instructions
On the Secondary NameNode (if used): sudo service
hadoop-hdfs-secondarynamenode stop

On each DataNode: sudo service


hadoop-hdfs-datanode stop

11

ZooKeeper

Stop HBase and HDFS


before stopping
ZooKeeper.

To stop the ZooKeeper server, use one of the following


commands on each ZooKeeper node:
sudo service zookeeper-server stop

or
sudo service zookeeper stop

Migrating Data between a CDH 4 and CDH 5 Cluster


You can migrate the data from a CDH 4 (or any Apache Hadoop) cluster to a CDH 5 cluster by using a tool that
copies out data in parallel, such as the DistCp tool offered in CDH 5. This can be useful if you are not planning
to upgrade your CDH 4 cluster itself at this point. The following sections provide information and instructions:

Requirements and Restrictions for Data Migration between CDH 4 and CDH 5 on page 72
Copying Data Between Two Clusters Using Distcp on page 73
Post-Migration Verification
Ports Used by DistCp

Requirements and Restrictions for Data Migration between CDH 4 and CDH 5
1. The CDH 5 cluster must have a MapReduce service running on it (MRv1 or YARN (MRv2)).
2. All the MapReduce nodes in the CDH 5 cluster should have full network access to all the nodes of the source
cluster. This allows you to perform the copy in a distributed manner.
3. To copy data between a secure and an insecure cluster, you must run the distcp command on the secure
cluster.
4. To copy data from a CDH 4 to a CDH 5 cluster, you can do one of the following:
Note:
The term source in this case refers to the CDH 4 (or other Hadoop) cluster you want to migrate or
copy data from; and destination refers to the CDH 5 cluster.
Running commands on the destination cluster, use the Hftp protocol for the source cluster, and HDFS
for the destination. (Hftp is read-only, so you must run DistCp on the destination cluster and pull the data
from the source cluster.) See Copying Data Between Two Clusters Using Distcp on page 73.
Note:
Do not use this method if one of the clusters is secure and the other is not.
Running commands on the source cluster, use the HDFS or webHDFS protocol for the source cluster, and
webHDFS for the destination. See Copying Data between a Secure and an Insecure Cluster using DistCp
and WebHDFS on page 76.
Running commands on the destination cluster, use webHDFS for the source cluster, and webHDFS for
the destination. See Copying Data between a Secure and an Insecure Cluster using DistCp and WebHDFS
on page 76.
The following restrictions currently apply (see Apache Hadoop Known Issues):
72 | Cloudera Administration

Managing CDH and Managed Services


DistCp does not work between a secure cluster and an insecure cluster in some cases.
As of CDH 5.1.3, DistCp does work between a secure and an insecure cluster if you use the webHDFS protocol
and run the command from the secure cluster side after setting
ipc.client.fallback-to-simple-auth-allowed to true, as described under Copying Data between a
Secure and an Insecure Cluster using DistCp and WebHDFS on page 76.
To use DistCp using Hftp from a secure cluster using SPNEGO, you must configure the dfs.https.port
property on the client to use the HTTP port (50070 by default).

Copying Data Between Two Clusters Using Distcp


Important: Do not run distcp as the HDFS user. The HDFS user is blacklisted for MapReduce jobs
by default.
You can use the distcp tool on the destination cluster to initiate the copy job to move the data. Between two
clusters running different versions of CDH, run the distcp tool with hftp:// as the source file system and
hdfs:// as the destination file system. This uses the HFTP protocol for the source and the HDFS protocol for
the destination. The default port for HFTP is 50070 and the default port for HDFS is 8020. Amazon S3 block and
native filesystems are also supported, via the s3a:// protocol.
Example of a source URI: hftp://namenode-location:50070/basePath
where namenode-location refers to the CDH 4 NameNode hostname as defined by its configured
fs.default.name and 50070 is the NameNode's HTTP server port, as defined by the configured
dfs.http.address.
Example of a destination URI: hdfs://nameservice-id/basePath or hdfs://namenode-location
This refers to the CDH 5 NameNode as defined by its configured fs.defaultFS.
The basePath in both the above URIs refers to the directory you want to copy, if one is specifically needed.
Example of an Amazon S3 Block Filesystem URI: s3a://bucket_name/path/to/file
S3 credentials can be provided in a configuration file (for example, core-site.xml):
<property>
<name>fs.s3a.access.key>
<value>...</value>
</property>
<property>
<name>fs.s3a.secret.key>
<value>...</value>
</property>

or on the command line as follows:


hadoop distcp -Dfs.s3a.access.key=... -Dfs.s3a.secret.key=... s3a://

The Distcp Command


For more help, and to see all the options available on the distcp command, use the following command to see
the built-in help:
$ hadoop distcp

Run the distcp copy by issuing a command such as the following on the CDH 5 cluster:

Cloudera Administration | 73

Managing CDH and Managed Services


Important: If you are using distcp as part of an upgrade, run the following commands on the
destination cluster only, in this example, the CDH 5 cluster, during the upgrade.
$ hadoop distcp hftp://cdh4-namenode:50070/ hdfs://CDH5-nameservice/
$ hadoop distcp s3a://bucket/ hdfs://CDH5-nameservice/

Or use a specific path, such as /hbase to move HBase data, for example:
$ hadoop distcp hftp://cdh4-namenode:50070/hbase hdfs://CDH5-nameservice/hbase
$ hadoop distcp s3a://bucket/file hdfs://CDH5-nameservice/bucket/file
distcp will then submit a regular MapReduce job that performs a file-by-file copy.
distcp is a general utility for copying files between distributed filesystems in different clusters. In general, you
can use distcp to copy files between compatible clusters in either direction. For instance, you can copy files
from a HDFS filesystem to an Amazon S3 filesystem. The examples given above are specific to using distcp to

assist with an upgrade from CDH 4 to CDH 5.


Protocol Support for Distcp
The following table lists support for using different protocols with the distcp command on different versions
of CDH. In the table, secure means that the cluster is configured to use Kerberos. Copying between a secure
cluster and an insecure cluster is only supported from CDH 5.1.3 onward, due to the inclusion of HDFS-6776.
Note: HFTP is a read-only protocol and cannot be used for the destination.
Source

Destination

Source
Destination
Where to
protocol and protocol and issue distcp
configuration configuration command

CDH 4

CDH 4

hdfs or
webhdfs,
insecure

hdfs or
webhdfs,
insecure

Source or
destination

ok

CDH 4

CDH 4

hftp, insecure hdfs or


webhdfs,
insecure

Destination

ok

CDH 4

CDH 4

hdfs or
webhdfs,
secure

hdfs or
webhdfs,
secure

Source or
destination

ok

CDH 4

CDH 4

hftp, secure

hdfs or
webhdfs,
secure

Destination

ok

CDH 4

CDH 5

webhdfs,
insecure

webhdfs or
Destination
hdfs, insecure

ok

CDH 4

CDH 5 ( 5.1.3
and newer)

webhdfs,
insecure

webhdfs,
secure

CDH 4

CDH 5

webhdfs or
webhdfs or
Destination
hftp, insecure hdfs, insecure

ok

CDH 4

CDH 5

webhdfs or
hftp, secure

webhdfs or
hdfs, secure

Destination

ok

CDH 4

CDH 5

hdfs or
webhdfs,
insecure

webhdfs,
insecure

Source

ok

74 | Cloudera Administration

Destination

Fallback
Status
Configuration
Required

yes

ok

Managing CDH and Managed Services


Source

Destination

Source
Destination
Where to
protocol and protocol and issue distcp
configuration configuration command

Fallback
Status
Configuration
Required

CDH 5

CDH 4

webhdfs ,
insecure

webhdfs,
insecure

Source or
destination

ok

CDH 5

CDH 4

webhdfs ,
insecure

hdfs, insecure Destination

ok

CDH 5

CDH 4

hdfs, insecure webhdfs,


insecure

Source

ok

CDH 5

CDH 4

hftp, insecure hdfs or


webhdfs,
insecure

Destination

ok

CDH 5

CDH 4

webhdfs,
secure

webhdfs,
secure

Source or
destination

ok

CDH 5

CDH 4

webhdfs or
hdfs, secure

webhdfs,
insecure

Source

ok

CDH 5

CDH 4

hdfs, secure

webhdfs,
secure

Source

ok

CDH 5

CDH 5

hdfs or
webhdfs,
insecure

hdfs or
webhdfs,
insecure

Source or
destination

ok

CDH 5

CDH 5

hftp, insecure hdfs or


webhdfs,
insecure

Destination

ok

CDH 5

CDH 5

hdfs or
webhdfs,
secure

hdfs or
webhdfs,
secure

Source or
destination

ok

CDH 5

CDH 5

hftp, secure

hdfs or
webhdfs,
secure

Destination

ok

CDH 5

CDH 5

hdfs or
webhdfs,
secure

webhdfs,
insecure

Source

yes

ok

CDH 5

CDH 5

hdfs or
webhdfs,
insecure

hdfs or
webhdfs,
secure

Destination

yes

ok

Note: distcp does not support hftp from an insecure cluster to a secure cluster.
To enable the fallback configuration, for copying between a secure cluster and an insecure one, add the following
to the HDFS core-default.xml, by using an advanced configuration snippet if you use Cloudera Manager, or
editing the file directly otherwise.
<property>
<name>ipc.client.fallback-to-simple-auth-allowed</name>
<value>true</value>
</property>

Cloudera Administration | 75

Managing CDH and Managed Services


Distcp between Secure Clusters in Distinct Kerberos Realms
This section briefly describes the steps required to copy data between two secure clusters in distinct Kerberos
realms.
Note: JDK version 1.7.x is required on both clusters when copying data between kerberized clusters
that are in different realms. For information about supported JDK versions, see Supported JDK Versions.

Specify the Destination Parameters in krb5.conf


Edit the krb5.conf file on the client (where the distcp job will be submitted) to include the destination hostname
and realm.
[realms]
HADOOP.QA.domain.COM = { kdc = kdc.domain.com:88 admin_server = admin.test.com:749
default_domain = domain.com supported_enctypes = arcfour-hmac:normal des-cbc-crc:normal
des-cbc-md5:normal des:normal des:v4 des:norealm des:onlyrealm des:afs3 }
[domain_realm]
.domain.com = HADOOP.test.domain.COM
domain.com = HADOOP.test.domain.COM
test03.domain.com = HADOOP.QA.domain.COM

(If TLS/SSL is enabled) Specify Truststore Properties


The following properties must be configured in the ssl-client.xml file on the client submitting the distcp job
to establish trust between the target and destination clusters.
<property>
<name>ssl.client.truststore.location</name>
<value>path_to_truststore</value>
</property>
<property>
<name>ssl.client.truststore.password</name>
<value>XXXXXX</value>
</property>
<property>
<name>ssl.client.truststore.type</name>
<value>jks</value>
</property>

Set HADOOP_CONF to the Destination Cluster


Set the HADOOP_CONF path to be the destination environment. If you are not using HFTP, set the HADOOP_CONF
path to the source environment instead.

Launch Distcp
Kinit on the client and launch the distcp job.
hadoop distcp hdfs://test01.domain.com:8020/user/alice
hdfs://test02.domain.com:8020/user/alice

If launching distcp fails, force Kerberos to use TCP instead of UDP by adding the following parameter to the
krb5.conf file on the client.
[libdefaults]
udp_preference_limit = 1

Copying Data between a Secure and an Insecure Cluster using DistCp and WebHDFS
You can use DistCp and WebHDFS to copy data between a secure cluster and an insecure cluster by doing the
following:

76 | Cloudera Administration

Managing CDH and Managed Services


1. Set ipc.client.fallback-to-simple-auth-allowed to true in core-site.xml on the secure cluster side:
<property>
<name>ipc.client.fallback-to-simple-auth-allowed</name>
<value>true</value>
</property>

2. Use commands such as the following from the secure cluster side only:
distcp webhdfs://insecureCluster webhdfs://secureCluster
distcp webhdfs://secureCluster webhdfs://insecureCluster

Post-migration Verification
After migrating data between the two clusters, it is a good idea to use hadoop fs -ls /basePath to verify
the permissions, ownership and other aspects of your files, and correct any problems before using the files in
your new cluster.

Managing Individual Services


The following sections cover the configuration and management of individual CDH and other services that have
specific and unique requirements or options.

Managing Flume
The Flume packages are installed by the Installation wizard, but the service is not created. This page documents
how to add, configure, and start the Flume service.
Adding a Flume Service
Required Role:
1.

On the Home > Status tab, click


to the right of the cluster name and select Add a Service. A list of service
types display. You can add one type of service at a time.
2. Select the Flume service and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

Cloudera Administration | 77

Managing CDH and Managed Services


IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
Configuring the Flume Agents
Required Role:
After you create a Flume service, you must first configure the agents before you start them. For detailed
information about Flume agent configuration, see the Flume User Guide.
The default Flume agent configuration provided in the Configuration File property of the Agent default role group
is a configuration for a single agent in a single tier; you should replace this with your own configuration. When
you add new agent roles, they are placed (initially) in the Agent default role group.
Agents that share the same configuration should be members of the same agent role group. You can create
new role groups and can move agents between them. If your Flume configuration has multiple tiers, you must
create an agent role group for each tier, and move each agent to be a member of the appropriate role group for
their tier.
A Flume agent role group Configuration File property can contain the configuration for multiple agents, since
each configuration property is prefixed by the agent name. You can set the agents' names using configuration
overrides to change the name of a specific agent without changing its role group membership. Different agents
can have the same name agent names do not have to be unique.
1. Go to the Flume service.
2. Click the Configuration tab.
3. Select Scope > Agent . Settings you make to the default role group apply to all agent instances unless you
associate those instances with a different role group, or override them for specific agents. See Modifying
Configuration Properties Using Cloudera Manager on page 10.
4. Set the Agent Name property to the name of the agent (or one of the agents) defined in the flume.conf
configuration file. The agent name can be comprised of letters, numbers, and the underscore character. You
can specify only one agent name here the name you specify will be used as the default for all Flume agent
instances, unless you override the name for specific agents. You can have multiple agents with the same
name they will share the configuration specified in on the configuration file.
5. Copy the contents of the flume.conf file, in its entirety, into the Configuration File property. Unless overridden
for specific agent instances, this property applies to all agents in the group. You can provide multiple agent
configurations in this file and use agent name overrides to specify which configuration to use for each agent.
Important: The name-value property pairs in the Configuration File property must include an
equal sign (=). For example, tier1.channels.channel1.capacity = 10000.

Setting a Flume Agent Name


Required Role:
If you have specified multiple agent configurations in a Flume agent role group Configuration File property, you
can set the agent name for an agent that uses a different configuration. Overriding the agent name will point
the agent to the appropriate properties specified in the agent configuration.
1.
2.
3.
4.
5.

Go to the Flume service.


Click the Configuration tab.
Select Scope > Agent.
Locate the Agent Name property or search for it by typing its name in the Search box.
Enter a name for the agent.

78 | Cloudera Administration

Managing CDH and Managed Services


If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
Using Flume with HDFS or HBase Sinks
If you want to use Flume with HDFS or HBase sinks, you can add a dependency to that service from the Flume
configuration page. This will automatically add the correct client configurations to the Flume agent's classpath.
Note: If you are using Flume with HBase, make sure that the /etc/zookeeper/conf/zoo.cfg file
either does not exist on the host of the Flume agent that is using an HBase sink, or that it contains
the correct ZooKeeper quorum.
Using Flume with Solr Sinks
Cloudera Manager provides a set of configuration settings under the Flume service to configure the Flume
Morphline Solr Sink. See Configuring the Flume Morphline Solr Sink for Use with the Solr Service on page 219 for
detailed instructions.
Updating Flume Agent Configurations
Required Role:
If you modify the Configuration File property after you have started the Flume service, update the configuration
across Flume agents as follows:
1. Go to the Flume service.
2. Select Actions > Update Config.

Managing HBase
Managing HBase
Cloudera Manager requires certain additional steps to set up and configure the HBase service.
Creating the HBase Root Directory
Required Role:
When adding the HBase service, the Add Service wizard automatically creates a root directory for HBase in HDFS.
If you quit the Add Service wizard or it does not finish, you can create the root directory outside the wizard by
doing these steps:
1. Choose Create Root Directory from the Actions menu in the HBase > Status tab.
2. Click Create Root Directory again to confirm.
Graceful Shutdown
Required Role:
A graceful shutdown of an HBase RegionServer allows the regions hosted by that RegionServer to be moved to
other RegionServers before stopping the RegionServer. Cloudera Manager provides the following configuration
options to perform a graceful shutdown of either an HBase RegionServer or the entire service.
To increase the speed of a rolling restart of the HBase service, set the Region Mover Threads property to a higher
value. This increases the number of regions that can be moved in parallel, but places additional strain on the
HMaster. In most cases, Region Mover Threads should be set to 5 or lower.

Cloudera Administration | 79

Managing CDH and Managed Services


Gracefully Shutting Down an HBase RegionServer
1.
2.
3.
4.
5.

Go to the HBase service.


Click the Instances tab.
From the list of Role Instances, select the RegionServer you want to shut down gracefully.
Select Actions for Selected > Decommission (Graceful Stop).
Cloudera Manager attempts to gracefully shut down the RegionServer for the interval configured in the
Graceful Shutdown Timeout configuration option, which defaults to 3 minutes. If the graceful shutdown fails,
Cloudera Manager forcibly stops the process by sending a SIGKILL (kill -9) signal. HBase will perform
recovery actions on regions that were on the forcibly stopped RegionServer.
6. If you cancel the graceful shutdown before the Graceful Shutdown Timeout expires, you can still manually
stop a RegionServer by selecting Actions for Selected > Stop, which sends a SIGTERM (kill -5) signal.
Gracefully Shutting Down the HBase Service
1. Go to the HBase service.
2. Select Actions > Stop. This tries to perform an HBase Master-driven graceful shutdown for the length of the
configured Graceful Shutdown Timeout (three minutes by default), after which it abruptly shuts down the
whole service.
Configuring the Graceful Shutdown Timeout Property
Required Role:
This timeout only affects a graceful shutdown of the entire HBase service, not individual RegionServers. Therefore,
if you have a large cluster with many RegionServers, you should strongly consider increasing the timeout from
its default of 180 seconds.
1.
2.
3.
4.
5.

Go to the HBase service.


Click the Configuration tab.
Select Scope > HBASE-1 (Service Wide)
Use the Search box to search for the Graceful Shutdown Timeout property and edit the value.
Click Save Changes to save this setting.

Adding the HBase Thrift Server


Required Role:
The Thrift Server role is not added by default when you install HBase. To add the Thrift Server role:
1.
2.
3.
4.

Go to the HBase service.


Click the Instances tab.
Click the Add Role Instances button.
Select the host(s) where you want to add the Thrift Server role (you only need one for Hue) and click Continue.
The Thrift Server role should appear in the instances list for the HBase server.
5. Select the Thrift Server role instance.
6. Select Actions for Selected > Start.
Enabling HBase Indexing
Required Role:
HBase indexing is dependent on the Key-Value Store Indexer service. The Key-Value Store Indexer service uses
the Lily HBase Indexer Service to index the stream of records being added to HBase tables. Indexing allows you
to query data stored in HBase with the Solr service.
1. Go to the HBase service.
2. Click the Configuration tab.

80 | Cloudera Administration

Managing CDH and Managed Services


3.
4.
5.
6.

Select Scope > HBASE-1 (Service Wide)


Select Category > Backup.
Select the Enable Replication and Enable Indexing properties.
Click Save Changes.

Adding a Custom Coprocessor


Required Role:
The HBase coprocessor framework provides a way to extend HBase with custom functionality. To configure
these properties in Cloudera Manager:
1.
2.
3.
4.
5.
6.

Select the HBase service.


Click the Configuration tab.
Select Scope > All.
Select Category > All.
Type HBase Coprocessor in the Search box.
You can configure the values of the following properties:
HBase Coprocessor Abort on Error (Service-Wide)
HBase Coprocessor Master Classes (Master Default Group)
HBase Coprocessor Region Classes (RegionServer Default Group)

7. Click Save Changes to commit the changes.


Enabling Hedged Reads on HBase
Required Role:
1.
2.
3.
4.
5.

Go to the HBase service.


Click the Configuration tab.
Select Scope > HBASE-1 (Service-Wide).
Select Category > Performance.
Configure the HDFS Hedged Read Threadpool Size and HDFS Hedged Read Delay Threshold properties. The
descriptions for each of these properties on the configuration pages provide more information.
6. Click Save Changes to commit the changes.
Advanced Configuration for Write-Heavy Workloads
HBase includes several advanced configuration parameters for adjusting the number of threads available to
service flushes and compactions in the presence of write-heavy workloads. Tuning these parameters incorrectly
can severely degrade performance and is not necessary for most HBase clusters. If you use Cloudera Manager,
configure these options using the HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml.
hbase.hstore.flusher.count

The number of threads available to flush writes from memory to disk. Never increase
hbase.hstore.flusher.count to more of 50% of the number of disks available to HBase. For example,
if you have 8 solid-state drives (SSDs), hbase.hstore.flusher.count should never exceed 4. This
allows scanners and compactions to proceed even in the presence of very high writes.
hbase.regionserver.thread.compaction.large and hbase.regionserver.thread.compaction.small
The number of threads available to handle small and large compactions, respectively. Never increase
either of these options to more than 50% of the number of disks available to HBase.
Ideally, hbase.regionserver.thread.compaction.small should be greater than or equal to
hbase.regionserver.thread.compaction.large, since the large compaction threads do more intense
work and will be in use longer for a given operation.

Cloudera Administration | 81

Managing CDH and Managed Services


In addition to the above, if you use compression on some column families, more CPU will be used when flushing
these column families to disk during flushes or compaction. The impact on CPU usage depends on the size of
the flush or the amount of data to be decompressed and compressed during compactions.

Starting and Stopping HBase


Use these instructions to start, stop, restart, rolling restart, or decommission HBase clusters or individual hosts.
Starting or Restarting HBase
You can start HBase hosts individually or as an entire cluster.
Starting or Restarting HBase Using Cloudera Manager
1. Go to the HBase service.
2. Click the Actions button and select Start.
3. To restart a running cluster, click Actions and select Restart or Rolling Restart. A rolling restart, which restarts
each RegionServer, one at a time, after a grace period. To configure the grace period, see Configuring the
Graceful Shutdown Timeout Property on page 80.
4. The Thrift service has no dependencies and can be restarted at any time. To stop or restart the Thrift service:

Go to the HBase service.


Select Instances.
Select the HBase Thrift Server instance.
Select Actions for Selected and select either Stop or Restart.

Starting or Restarting HBase Using the Command Line


Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
If you need the ability to perform a rolling restart, Cloudera recommends managing your cluster with Cloudera
Manager.
1. To start a HBase cluster using the command line, start the HBase Master by using the sudo hbase-master
start command on RHEL or SuSE, or the sudo hadoop-hbase-regionserver start command on Ubuntu
or Debian. The HMaster starts the RegionServers automatically.
2. To start a RegionServer manually, use the sudo hbase-regionserver start command on RHEL or SuSE,
or the sudo hadoop-hbase-regionserver start command on Ubuntu or Debian. Running multiple
RegionServer processes on the same host is not supported.
3. The Thrift service has no dependencies and can be restarted at any time. To start the Thrift server, use the
hbase-thrift start on RHEL or SuSE, or the hadoop-hbase-thrift start on Ubuntu or Debian.
Stopping HBase
You can stop a single HBase host, all hosts of a given type, or all hosts in the cluster.
Stopping HBase Using Cloudera Manager
1. To stop or decommission a single RegionServer:
a. Go to the HBase service.
b. Click the Instances tab.
c. From the list of Role Instances, select the RegionServer or RegionServers you want to stop or
decommission.

82 | Cloudera Administration

Managing CDH and Managed Services


d. Select Actions for Selected and select either Decommission (Graceful Stop) or Stop.
Graceful Stop causes the regions to be redistributed to other RegionServers, increasing availability
during the RegionServer outage. Cloudera Manager waits for an interval determined by the Graceful
Shutdown timeout interval, which defaults to three minutes. If the graceful stop does not succeed
within this interval, the RegionServer is stopped with a SIGKILL (kill -9) signal. Recovery will be
initiated on affected regions.
Stop happens immediately and does not redistribute the regions. It issues a SIGTERM (kill -5) signal.
2. To stop or decommission a single HMaster, select the Master and go through the same steps as above.
3. To stop or decommission the entire cluster, select the Actions button at the top of the screen (not Actions
for selected) and select Decommission (Graceful Stop) or Stop.
Stopping HBase Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Shut down the Thrift server by using the hbase-thrift stop command on the Thrift server host. sudo
service hbase-thrift stop

2. Shut down each RegionServer by using the hadoop-hbase-regionserver stop command on the
RegionServer host.
sudo service hadoop-hbase-regionserver stop

3. Shut down backup HMasters, followed by the main HMaster, by using the hbase-master stop command.
sudo service hbase-master stop

Configuring the HBase Canary


The HBase canary is an optional service that periodically checks that a RegionServer is alive. This canary is
different from the Cloudera Service Monitoring canary and is provided by the HBase service. The HBase canary
is disabled by default. After enabling the canary, you can configure several different thresholds and intervals
relating to it, as well as exclude certain tables from the canary checks. The canary works on Kerberos-enabled
clusters if you have the HBase client configured to use Kerberos.
Configure the HBase Canary Using Cloudera Manager
Required Role:
1.
2.
3.
4.
5.

Go to the HBase service.


Click the Configuration tab.
Select Scope > HBase or HBase Service-Wide.
Select Category > Monitoring.
Locate the HBase Canary property or search for it by typing its name in the Search box. Several properties
have Canary in the property name.
6. Select the checkbox.
7. Review other HBase Canary properties to configure the specific behavior of the canary.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

Cloudera Administration | 83

Managing CDH and Managed Services


8. Click Save Changes to commit the changes.
9. Restart the role.
10. Restart the service.
Configure the HBase Canary Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
The HBase canary is a Java class. To run it from the command line, in the foreground, issue a command similar
to the following, as the HBase user:
$ /usr/bin/hbase org.apache.hadoop.hbase.tool.Canary

To start the canary in the background, add the --daemon option. You can also use this option in your HBase
startup scripts.
$ /usr/bin/hbase org.apache.hadoop.hbase.tool.Canary --daemon

The canary has many options. To see usage instructions, add the --help parameter:
$ /usr/bin/hbase org.apache.hadoop.hbase.tool.Canary --help

Checking and Repairing HBase Tables


HBaseFsck (hbck) is a command-line tool that checks for region consistency and table integrity problems and
repairs corruption. It works in two basic modes a read-only inconsistency identifying mode and a multi-phase
read-write repair mode.
Read-only inconsistency identification: In this mode, which is the default, a report is generated but no repairs
are attempted.
Read-write repair mode: In this mode, if errors are found, hbck attempts to repair them.
You can run hbck manually or configure the hbck poller to run hbck periodically.
Always run HBase administrative commands such as the HBase Shell, hbck, or bulk-load commands as the
HBase user (typically hbase).
Running hbck Manually
The hbck command is located in the bin directory of the HBase install.
With no arguments, hbck checks HBase for inconsistencies and prints OK if no inconsistencies are found, or
the number of inconsistencies otherwise.
With the -details argument, hbck checks HBase for inconsistencies and prints a detailed report.
To limit hbck to only checking specific tables, provide them as a space-separated list: hbck <table1>
<table2>

Warning: The following hbck options modify HBase metadata and are dangerous. They are not
coordinated by the HMaster and can cause further corruption by conflicting with commands that are
currently in progress or coordinated by the HMaster. Even if the HMaster is down, it may try to recover
the latest operation when it restarts. These options should only be used as a last resort. The hbck
command can only fix actual HBase metadata corruption and is not a general-purpose maintenance
tool. Before running these commands, consider contacting Cloudera Support for guidance. In addition,
running any of these commands requires an HMaster restart.

84 | Cloudera Administration

Managing CDH and Managed Services


If region-level inconsistencies are found, use the -fix argument to direct hbck to try to fix them. The following
sequence of steps is followed:
1. The standard check for inconsistencies is run.
2. If needed, repairs are made to tables.
3. If needed, repairs are made to regions. Regions are closed during repair.
You can also fix individual region-level inconsistencies separately, rather than fixing them automatically with
the -fix argument.
-fixAssignments repairs unassigned, incorrectly assigned or multiply assigned regions.
-fixMeta removes rows from hbase:meta when their corresponding regions are not present in HDFS
and adds new meta rows if regions are present in HDFS but not in hbase:meta.
-repairHoles creates HFiles for new empty regions on the filesystem and ensures that the new regions
are consistent.
-fixHdfsOrphans repairs a region directory that is missing a region metadata file (the .regioninfo
file).
-fixHdfsOverlaps fixes overlapping regions. You can further tune this argument using the following
options:
-maxMerge <n> controls the maximum number of regions to merge.
-sidelineBigOverlaps attempts to sideline the regions which overlap the largest number of other
regions.
-maxOverlapsToSideline <n> limits the maximum number of regions to sideline.
To try to repair all inconsistencies and corruption at once, use the -repair option, which includes all the
region and table consistency options.
For more details about the hbck command, see Appendix C of the HBase Reference Guide.

Hedged Reads
Hadoop 2.4 introduced a new feature called hedged reads. If a read from a block is slow, the HDFS client starts
up another parallel, 'hedged' read against a different block replica. The result of whichever read returns first is
used, and the outstanding read is cancelled. This feature helps in situations where a read occasionally takes a
long time rather than when there is a systemic problem. Hedged reads can be enabled for HBase when the
HFiles are stored in HDFS. This feature is disabled by default.
Enabling Hedged Reads for HBase Using Cloudera Manager
Required Role:
1.
2.
3.
4.
5.

Go to the HBase service.


Click the Configuration tab.
Select Scope > HBASE-1 (Service-Wide).
Select Category > Performance.
Configure the HDFS Hedged Read Threadpool Size and HDFS Hedged Read Delay Threshold properties. The
descriptions for each of these properties on the configuration pages provide more information.
6. Click Save Changes to commit the changes.
Enabling Hedged Reads for HBase Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.

Cloudera Administration | 85

Managing CDH and Managed Services


To enable hedged reads for HBase, edit the hbase-site.xml file on each server. Set
dfs.client.hedged.read.threadpool.size to the number of threads to dedicate to running hedged threads,
and set the dfs.client.hedged.read.threshold.millis configuration property to the number of milliseconds
to wait before starting a second read against a different block replica. Set
dfs.client.hedged.read.threadpool.size to 0 or remove it from the configuration to disable the feature.
After changing these properties, restart your cluster.
The following is an example configuration for hedged reads for HBase.
<property>
<name>dfs.client.hedged.read.threadpool.size</name>
<value>20</value> <!-- 20 threads -->
</property>
<property>
<name>dfs.client.hedged.read.threshold.millis</name>
<value>10</value> <!-- 10 milliseconds -->
</property>

Monitoring the Performance of Hedged Reads


You can monitor the performance of hedged reads using the following metrics emitted by Hadoop when hedged
reads are enabled.
hedgedReadOps - the number of hedged reads that have occurred
hedgeReadOpsWin - the number of times the hedged read returned faster than the original read

Configuring the Blocksize for HBase


The blocksize is an important configuration option for HBase. HBase data is stored in one (after a major
compaction) or more (possibly before a major compaction) HFiles per column family per region. It determines
both of the following:
The blocksize for a given column family determines the smallest unit of data HBase can read from the column
family's HFiles.
It is also the basic unit of measure cached by a RegionServer in the BlockCache.
The default blocksize is 64 KB. The appropriate blocksize is dependent upon your data and usage patterns. Use
the following guidelines to tune the blocksize size, in combination with testing and benchmarking as appropriate.
Warning: The default blocksize is appropriate for a wide range of data usage patterns, and tuning
the blocksize is an advanced operation. The wrong configuration can negatively impact performance.
Consider the average key/value size for the column family when tuning the blocksize. You can find the
average key/value size using the HFile utility:
$ hbase org.apache.hadoop.hbase.io.hfile.HFile -f /path/to/HFILE -m -v
...
Block index size as per heapsize: 296
reader=hdfs://srv1.example.com:9000/path/to/HFILE, \
compression=none, inMemory=false, \
firstKey=US6683275_20040127/mimetype:/1251853756871/Put, \
lastKey=US6684814_20040203/mimetype:/1251864683374/Put, \
avgKeyLen=37, avgValueLen=8, \
entries=1554, length=84447
...

Consider the pattern of reads to the table or column family. For instance, if it is common to scan for 500 rows
on various parts of the table, performance might be increased if the blocksize is large enough to encompass
500-1000 rows, so that often, only one read operation on the HFile is required. If your typical scan size is
only 3 rows, returning 500-1000 rows would be overkill.
It is difficult to predict the size of a row before it is written, because the data will be compressed when it is
written to the HFile. Perform testing to determine the correct blocksize for your data.

86 | Cloudera Administration

Managing CDH and Managed Services


Configuring the Blocksize for a Column Family
You can configure the blocksize of a column family at table creation or by disabling and altering an existing table.
These instructions are valid whether or not you use Cloudera Manager to manage your cluster.
hbase>
hbase>
hbase>
hbase>

create test_table,{NAME => test_cf, BLOCKSIZE => '262144'}


disable 'test_table'
alter 'test_table', {NAME => 'test_cf', BLOCKSIZE => '524288'}
enable 'test_table'

After changing the blocksize, the HFiles will be rewritten during the next major compaction. To trigger a major
compaction, issue the following command in HBase Shell.
hbase> major_compact 'test_table'

Depending on the size of the table, the major compaction can take some time and have a performance impact
while it is running.
Monitoring Blocksize Metrics
Several metrics are exposed for monitoring the blocksize by monitoring the blockcache itself.

Configuring the HBase BlockCache


In the default configuration, HBase uses a single on-heap cache. If you configure the off-heap BucketCache,
the on-heap cache is used for Bloom filters and indexes,and the off-heap BucketCache is used to cache data
blocks. This is referred to as the Combined Blockcache configuration. The Combined BlockCache allows you to
use a larger in-memory cache while reducing the negative impact of garbage collection in the heap, because
HBase manages the BucketCache, rather than relying on the garbage collector.
Contents of the BlockCache
In order to size the BlockCache correctly, you need to understand what HBase places into it.
Your data: Each time a Get or Scan operation occurs, the result is added to the BlockCache if it was not already
cached there. If you use the BucketCache, data blocks are always cached in the BucketCache.
Row keys: When a value is loaded into the cache, its row key is also cached. This is one reason that it is
important to make your row keys as small as possible. A larger row key takes up more space in the cache.
hbase:meta: The hbase:meta catalog table keeps track of which RegionServer is serving which regions. It
can consume several megabytes of cache if you have a large number of regions. It is given in-memory access
priority, which means HBase attempts to keep it in the cache as long as possible.
Indexes of HFiles: HBase stores its data in HDFS in a format called HFile. These HFiles contain indexes which
allow HBase to seek for data within them without needing to open the entire HFile. The size of an index is a
factor of the block size, the size of your row keys, and the amount of data you are storing. For big data sets,
the size can exceed 1 GB per RegionServer, although it is unlikely that the entire index will be in the cache at
the same time. If you use the BucketCache, indexes are always cached on-heap.
Bloom filters: If you use Bloom filters, they are stored in the BlockCache. If you use the BucketCache, Bloom
filters are always cached on-heap.
The sum of the sizes of these objects is highly dependent on your usage patterns and the characteristics of your
data. For this reason, the HBase Web UI and Cloudera Manager each expose several metrics to help you size
and tune the BlockCache.
Deciding Whether To Use the BucketCache
The HBase team has published the results of exhaustive BlockCache testing, which revealed the following
guidelines.
If the result of a Get or Scan typically fits completely in the heap, the default configuration, which uses the
on-heap LruBlockCache, is the best choice, as the L2 cache will not provide much benefit. If the eviction
rate is low, garbage collection can be 50% less than that of the BucketCache, and throughput can be at least
20% higher.
Cloudera Administration | 87

Managing CDH and Managed Services


Otherwise, if your cache is experiencing a consistently high eviction rate, use the BucketCache, which causes
30-50% of the garbage collection of LruBlockCache when the eviction rate is high.
BucketCache using file mode on solid-state disks has a better garbage-collection profile but lower throughput
than BucketCache using off-heap memory.
Bypassing the BlockCache
If the data needed for a specific but atypical operation does not all fit in memory, using the BlockCache can be
counter-productive, because data that you are still using may be evicted, or even if other data is not evicted,
excess garbage collection can adversely effect performance. For this type of operation, you may decide to bypass
the BlockCache. To bypass the BlockCache for a given Scan or Get, use the setCacheBlocks(false) method.
In addition, you can prevent a specific column family's contents from being cached, by setting its BLOCKCACHE
configuration to false. Use the following syntax in HBase Shell:
hbase> alter 'myTable', 'myCF', CONFIGURATION => {BLOCKCACHE => 'false'}

Cache Eviction Priorities


Both the on-heap cache and the off-heap BucketCache use the same cache priority mechanism to decide which
cache objects to evict in order to make room for new objects. Three levels of block priority allow for scan-resistance
and in-memory column families. Objects evicted from the cache are subject to garbage collection.
Single access priority: The first time a block is loaded from HDFS, it is given single access priority, which
means that it will be part of the first group to be considered during evictions. Scanned blocks are more likely
to be evicted than blocks that are used more frequently.
Multi access priority: If a block in the single access priority group is accessed again, it is assigned multi access
priority, which means that it is part of the second group considered during evictions, and is therefore less
likely to be evicted.
In-memory access priority: If the block belongs to a column family which is configured with the in-memory
configuration option, it is assigned the in memory access priority, regardless of its access pattern. This group
is the last group considered during evictions, but is not guaranteed not to be evicted. Catalog tables are
configured with in-memory access priority.
To configure a column family for in-memory access, use the following syntax in HBase Shell:
hbase> alter 'myTable', 'myCF', CONFIGURATION => {IN_MEMORY => 'true'}

To use the Java API to configure a column family for in-memory access, use the
HColumnDescriptor.setInMemory(true) method.
Sizing the BlockCache
When you use the LruBlockCache, the blocks needed to satisfy each read are cached, evicting older cached
objects if the LruBlockCache is full. The size cached objects for a given read may be significantly larger than
the actual result of the read. For instance, if HBase needs to scan through 20 HFile blocks to return a 100 byte
result, and the HFile blocksize is 100 KB, the read will add 20 * 100 KB to the LruBlockCache.
Because the LruBlockCache resides entirely within the Java heap, it is important to understand how much
heap is available to HBase and what percentage of the heap is available to the LruBlockCache. By default, the
amount of HBase's heap reserved for the LruBlockCache (hfile.block.cache.size) is .25, or 25%. To
determine the amount of heap available for the LruBlockCache, use the following formula. The 0.99 factor
allows 1% of heap to be available as a "working area" for evicting items from the cache. If you use the BucketCache,
the on-heap LruBlockCache only stores indexes and Bloom filters, and data blocks are cached in the off-heap
BucketCache.
number of RegionServers * heap size * hfile.block.cache.size * 0.99

To tune the size of the LruBlockCache, you can add RegionServers or increase the total Java heap on a given
RegionServer to increase it, or you can tune hfile.block.cache.size to reduce it. Reducing it will cause cache

88 | Cloudera Administration

Managing CDH and Managed Services


evictions to happen more often, but will reduce the time it takes to perform a cycle of garbage collection. Increasing
the heap will cause garbage collection to take longer but happen less frequently.
About the off-heap BucketCache
If the BucketCache is enabled, it stores data blocks, leaving the on-heap cache free for storing indexes and Bloom
filters. The physical location of the BucketCache storage can be either in memory (off-heap) or in a file stored
in a fast disk.
Off-heap: This is the default configuration.
File-based: You can use the file-based storage mode to store the BucketCache on an SSD or FusionIO device,
Starting in CDH 5.4 (HBase 1.0), it is possible to configure a column family to keep its data blocks in the L1 cache
instead of the BucketCache, using the HColumnDescriptor.cacheDataInL1(true) method or by using the
following syntax in HBase Shell:
hbase> alter 'myTable', CONFIGURATION => {CACHE_DATA_IN_L1 => 'true'}}

Configuring the off-heap BucketCache


This table summaries the important configuration properties for the BucketCache. To configure the BucketCache,
see Configuring the off-heap BucketCache Using Cloudera Manager on page 90 or Configuring the off-heap
BucketCache Using the Command Line on page 90
Table 1: BucketCache Configuration Properties
Property

Default

Description

hbase.bucketcache.combinedcache.enabled true

When BucketCache is enabled, use


it as a L2 cache for LruBlockCache.
If set to true, indexes and Bloom
filters are kept in the
LruBlockCache and the data blocks
are kept in the BucketCache

hbase.bucketcache.ioengine

none (BucketCache is disabled by


default)

Where to store the contents of the


BucketCache. Either onheap or
file:/path/to/file.

hfile.block.cache.size

0.4

A float between 0.0 and 1.0. This


factor multiplied by the Java heap
size is the size of the L1 cache. In
other words, the percentage of the
Java heap to use for the L1 cache.

hbase.bucketcache.size

64 MB, which is the default blocksize The total size of the BucketCache,
in megabytes. The size you configure
depends on the amount of memory
available to HBase, or the size of
your FusionIO device or SSD.

hbase.bucketcache.sizes

not set

A comma-separated list of sizes for


buckets for the BucketCache if you
prefer to use multiple sizes. The
sizes should be multiples of the
default blocksize, ordered from
smallest to largest. The sizes you
use will depend on your data
patterns. This parameter is
experimental.

Cloudera Administration | 89

Managing CDH and Managed Services


Property

Default

Description

-XX:MaxDirectMemorySize

not set

A JVM option to configure the


maximum amount of direct memory
available to the JVM. This value
should be larger than the
combination of the heap and
off-heap BucketCache.

Configuring the off-heap BucketCache Using Cloudera Manager


1. Go to the HBase service.
2. Click the Configuration tab.
3. Edit the parameter HBASE_OFFHEAPSIZE in the HBase Service Advanced Configuration Snippet for
hbase-env.shand set it to a value (such as 5G) which will accommodate your desired L2 cache size, in addition
to space reserved for cache management.
4. Edit the parameter HBASE_OPTS in the HBase Service Advanced Configuration Snippet for hbase-env.sh and
add the JVM option -XX:MaxDirectMemorySize=<size>G, replacing <size> with a value large enough to
contain your heap and off-heap BucketCache, expressed as a number of gigabytes.
5. Add the following settings to the HBase Service Advanced Configuration Snippet for hbase-site.xml, using
values appropriate to your situation. See Table 1: BucketCache Configuration Properties on page 89.
<property>
<name>hbase.bucketcache.ioengine</name>
<value>offheap</value>
</property>
<property>
<name>hfile.block.cache.size</name>
<value>0.2</value>
</property>
<property>
<name>hbase.bucketcache.size</name>
<value>4096</value>
</property>

6. Click Save Changes to commit the changes.


7. Restart or rolling restart your RegionServers for the changes to take effect.
Configuring the off-heap BucketCache Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. First, verify the RegionServer's off-heap size, and if necessary, tune it by editing the hbase-env.sh file and
adding a line like the following:
HBASE_OFFHEAPSIZE=5G

Set it to a value which will accommodate your desired L2 cache size, in addition to space reserved for cache
management.
2. Edit the parameter HBASE_OPTS in the hbase-env.sh file and add the JVM option
-XX:MaxDirectMemorySize=<size>G, replacing <size> with a value large enough to contain your heap
and off-heap BucketCache, expressed as a number of gigabytes.

90 | Cloudera Administration

Managing CDH and Managed Services


3. Next, configure the properties in Table 1: BucketCache Configuration Properties on page 89 as appropriate,
using the example below as a model.
<property>
<name>hbase.bucketcache.ioengine</name>
<value>offheap</value>
</property>
<property>
<name>hfile.block.cache.size</name>
<value>0.2</value>
</property>
<property>
<name>hbase.bucketcache.size</name>
<value>4194304</value>
</property>

4. Restart each RegionServer for the changes to take effect.

Monitoring the BlockCache


Cloudera Manager provides metrics to monitor the performance of the BlockCache, to assist you in tuning your
configuration.
You can view further detail and graphs using the RegionServer UI. To access the RegionServer UI in Cloudera
Manager, go to the Cloudera Manager page for the host, click the RegionServer process, and click HBase
RegionServer Web UI.
If you do not use Cloudera Manager, access the BlockCache reports at
http://regionServer_host:22102/rs-status#memoryStats, replacing regionServer_host with the
hostname or IP address of your RegionServer.

Configuring the HBase Scanner Heartbeat


A scanner heartbeat check enforces a time limit on the execution of scan RPC requests. This helps prevent scans
from taking too long and causing a timeout at the client.
When the server receives a scan RPC request, a time limit is calculated to be half of the smaller of two values:
hbase.client.scanner.timeout.period and hbase.rpc.timeout (which both default to 60000 milliseconds,
or one minute). When the time limit is reached, the server returns the results it has accumulated up to that
point. This result set may be empty. If your usage pattern includes that scans will take longer than a minute,
you can increase these values.
To make sure the timeout period is not too short, you can configure
hbase.cells.scanned.per.heartbeat.check to a minimum number of cells that must be scanned before
a timeout check occurs. The default value is 10000. A smaller value causes timeout checks to occur more often.
Configure the Scanner Heartbeat Using Cloudera Manager
Required Role:
1.
2.
3.
4.
5.
6.
7.

Go to the HBase service.


Click the Configuration tab.
Select HBase or HBase Service-Wide.
Select Category > Main.
Locate the RPC Timeout property or search for it by typing its name in the Search box.
Edit the property.
To modify the default values for hbase.client.scanner.timeout.period or
hbase.cells.scanned.per.heartbeat.check, search for HBase Service Advanced Configuration Snippet
(Safety Valve) for hbase-site.xml. Paste one or both of the following properties into the field and modify the
values as needed.
<property>
<name>hbase.client.scanner.timeout.period</name>

Cloudera Administration | 91

Managing CDH and Managed Services


<value>60000</value>
</property>
<property>
<name>hbase.cells.scanned.per.heartbeat.check</name>
<value>10000</value>
</property>

8. Click Save Changes to commit the changes.


9. Restart the role.
10. Restart the service.
Configure the Scanner Heartbeat Using the Command Line
1. Edit hbase-site.xml and add the following properties, modifying the values as needed.
<property>
<name>hbase.rpc.timeout</name>
<value>60000</value>
</property>
<property>
<name>hbase.client.scanner.timeout.period</name>
<value>60000</value>
</property>
<property>
<name>hbase.cells.scanned.per.heartbeat.check</name>
<value>10000</value>
</property>

2. Distribute the modified hbase-site.xml to all your cluster nodes and restart the HBase master and
RegionServer processes for the change to take effect.

Limiting the Speed of Compactions


You can limit the speed at which HBase compactions run, by configuring
hbase.regionserver.throughput.controller and its related settings. The default controller is
org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController,

which uses the following algorithm:


1. If compaction pressure is greater than 1.0, there is no speed limitation.
2. In off-peak hours, use a fixed throughput limitation, configured in
hbase.hstore.compaction.throughput.offpeak.
3. In normal hours, the max throughput is tuned between
hbase.hstore.compaction.throughput.higher.bound and
hbase.hstore.compaction.throughput.lower.bound (which default to 20 MB/sec and 10 MB/sec
respectively), using the following formula, where compactionPressure is between 0.0 and 1.0. The
compactionPressure refers to the number of store files that require compaction.
lower + (higer - lower) * compactionPressure

To disable compaction speed limits, set hbase.regionserver.throughput.controller to


org.apache.hadoop.hbase.regionserver.compactions.NoLimitCompactionThroughputController.

Configure the Compaction Speed Using Cloudera Manager


Required Role:
1.
2.
3.
4.

Go to the HBase service.


Click the Configuration tab.
Select HBase or HBase Service-Wide.
Search for HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml. Paste the relevant
properties into the field and modify the values as needed. See Configure the Compaction Speed Using the
Command Line on page 93 for an explanation of the properties.

92 | Cloudera Administration

Managing CDH and Managed Services


5. Click Save Changes to commit the changes.
6. Restart the role.
7. Restart the service.
Configure the Compaction Speed Using the Command Line
1. Edit hbase-site.xml and add the relevant properties, modifying the values as needed. Default values are
shown.
<property>
<name>hbase.hstore.compaction.throughput.controller</name>
<value>org.apache.hadoop.hbase.regionserver.compactions.PressureAwareCompactionThroughputController</value>
</property>
<property>
<name>hbase.hstore.compaction.throughput.higher.bound</name>
<value>20971520</value>
<description>The default is 20 MB/sec</description>
</property>
<property>
<name>hbase.hstore.compaction.throughput.lower.bound</name>
<value>10485760</value>
<description>The default is 10 MB/sec</description>
</property>
<property>
<name>hbase.hstore.compaction.throughput.offpeak</name>
<value>9223372036854775807</value>
<description>The default is Long.MAX_VALUE, which effectively means no
limitation</description>
</property>
<property>
<name>hbase.hstore.compaction.throughput.tune.period</name>
<value>60000</value>
<description>
</property>

2. Distribute the modified hbase-site.xml to all your cluster nodes and restart the HBase master and
RegionServer processes for the change to take effect.

Reading Data from HBase


Get and Scan are the two ways to read data from HBase, aside from manually parsing HFiles. A Get is simply
a Scan limited by the API to one row. A Scan fetches zero or more rows of a table. By default, a Scan reads the
entire table from start to end. You can limit your Scan results in several different ways, which affect the Scan's

load in terms of IO, network, or both, as well as processing load on the client side. This topic is provided as a
quick reference. Refer to the API documentation for Scan for more in-depth information. You can also perform
Gets and Scan using the HBase Shell.
Specify a startrow and/or stoprow. Neither startrow nor stoprow need to exist. Because HBase sorts
rows lexicographically, it will return the first row after startrow would have occurred, and will stop returning
rows after stoprow would have occurred.The goal is to reduce IO and network.
The startrow is inclusive and the stoprow is exclusive. Given a table with rows a, b, c, d, e, f, and startrow
of c and stoprow of f, rows c-e are returned.
If you omit startrow, the first row of the table is the startrow.
If you omit the stoprow, all results after startrow (including startrow) are returned.
If startrow is lexicographically after stoprow, and you set Scan setReversed(boolean reversed) to
true, the results are returned in reverse order. Given the same table above, with rows a-f, if you specify
c as the stoprow and f as the startrow, rows f, e, and d are returned.
Scan()
Scan(byte[] startRow)
Scan(byte[] startRow, byte[] stopRow)

Cloudera Administration | 93

Managing CDH and Managed Services


Specify a scanner cache that will be filled before the Scan result is returned, setting setCaching to the
number of rows to cache before returning the result. By default, the caching setting on the table is used. The
goal is to balance IO and network load.
public Scan setCaching(int caching)

To limit the number of columns if your table has very wide rows (rows with a large number of columns), use
setBatch(int batch) and set it to the number of columns you want to return in one batch. A large number of
columns is not a recommended design pattern.
public Scan setBatch(int batch)

To specify a maximum result size, use setMaxResultSize(long), with the number of bytes. The goal is to
reduce IO and network.
public Scan setMaxResultSize(long maxResultSize)

When you use setCaching and setMaxResultSize together, single server requests are limited by either
number of rows or maximum result size, whichever limit comes first.
You can limit the scan to specific column families or columns by using addFamily or addColumn. The goal
is to reduce IO and network. IO is reduced because each column family is represented by a Store on each
region server, and only the Stores representing the specific column families in question need to be accessed.
public Scan addColumn(byte[] family,
byte[] qualifier)
public Scan addFamily(byte[] family)

You can specify a range of timestamps or a single timestamp by specifying setTimeRange or setTimestamp.
public Scan setTimeRange(long minStamp,
long maxStamp)
throws IOException
public Scan setTimeStamp(long timestamp)
throws IOException

You can retrieve a maximum number of versions by using setMaxVersions.


public Scan setMaxVersions(int maxVersions)

You can use a filter by using setFilter. Filters are discussed in detail in HBase Filtering on page 95 and
the Filter API.
public Scan setFilter(Filter filter)

You can disable the server-side block cache for a specific scan by executing setCacheBlocks(boolean).
This is an expert setting and should only be used if you know what you are doing.
Perform Scans Using HBase Shell
You can perform scans using HBase Shell, for testing or quick queries. Use the following guidelines or issue the
scan command in HBase Shell with no parameters for more usage information. This represents only a subset
of possibilities.
# Display usage information
hbase> scan
# Scan all rows of table 't1'
hbase> scan 't1'

94 | Cloudera Administration

Managing CDH and Managed Services


# Specify a startrow, limit the result to 10 rows, and only return selected columns
hbase> scan 't1', {COLUMNS => ['c1', 'c2'], LIMIT => 10, STARTROW => 'xyz'}
# Specify a timerange
hbase> scan 't1', {TIMERANGE => [1303668804, 1303668904]}
# Specify a custom filter
hbase> scan 't1', {FILTER => org.apache.hadoop.hbase.filter.ColumnPaginationFilter.new(1,
0)}
# Disable the block cache for a specific scan (experts only)
hbase> scan 't1', {COLUMNS => ['c1', 'c2'], CACHE_BLOCKS => false}

Hedged Reads
Hadoop 2.4 introduced a new feature called hedged reads. If a read from a block is slow, the HDFS client starts
up another parallel, 'hedged' read against a different block replica. The result of whichever read returns first is
used, and the outstanding read is cancelled. This feature helps in situations where a read occasionally takes a
long time rather than when there is a systemic problem. Hedged reads can be enabled for HBase when the
HFiles are stored in HDFS. This feature is disabled by default.
Enabling Hedged Reads for HBase Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
To enable hedged reads for HBase, edit the hbase-site.xml file on each server. Set
dfs.client.hedged.read.threadpool.size to the number of threads to dedicate to running hedged threads,
and set the dfs.client.hedged.read.threshold.millis configuration property to the number of milliseconds
to wait before starting a second read against a different block replica. Set
dfs.client.hedged.read.threadpool.size to 0 or remove it from the configuration to disable the feature.
After changing these properties, restart your cluster.
The following is an example configuration for hedged reads for HBase.
<property>
<name>dfs.client.hedged.read.threadpool.size</name>
<value>20</value> <!-- 20 threads -->
</property>
<property>
<name>dfs.client.hedged.read.threshold.millis</name>
<value>10</value> <!-- 10 milliseconds -->
</property>

HBase Filtering
When reading data from HBase using Get or Scan operations, you can use custom filters to return a subset of
results to the client. While this does not reduce server-side IO, it does reduce network bandwidth and reduces
the amount of data the client needs to process. Filters are generally used via the Java API, but can be used from
HBase Shell for testing and debugging purposes.
For more information on Gets and Scans in HBase, see Reading Data from HBase on page 93.
Filter Syntax Guidelines
HBase filters take zero or more arguments, in parentheses. Where the argument is a string, it is surrounded by
single quotes ('string').

Cloudera Administration | 95

Managing CDH and Managed Services


Logical Operators, Comparison Operators and Comparators
Filters can be combined together with logical operators. Some filters take a combination of comparison operators
and comparators. Following is the list of each.
Logical Operators

AND - the key-value must pass both the filters to be included in the results.
OR - the key-value must pass at least one of the filters to be included in the results.
SKIP - for a particular row, if any of the key-values do not pass the filter condition, the entire row is skipped.
WHILE - For a particular row, it continues to emit key-values until a key-value is reached that fails the filter
condition.
Compound Filters - Using these operators, a hierarchy of filters can be created. For example:
(Filter1 AND Filter2)OR(Filter3 AND Filter4)

Comparison Operators

LESS (<)
LESS_OR_EQUAL (<=)
EQUAL (=)
NOT_EQUAL (!=)
GREATER_OR_EQUAL (>=)
GREATER (>)
NO_OP (no operation)

Comparators
BinaryComparator - lexicographically compares against the specified byte array using the
Bytes.compareTo(byte[], byte[]) method.
BinaryPrefixComparator - lexicographically compares against a specified byte array. It only compares up to
the length of this byte array.
RegexStringComparator - compares against the specified byte array using the given regular expression. Only
EQUAL and NOT_EQUAL comparisons are valid with this comparator.
SubStringComparator - tests whether or not the given substring appears in a specified byte array. The
comparison is case insensitive. Only EQUAL and NOT_EQUAL comparisons are valid with this comparator.
Examples
Example1: >, 'binary:abc' will match everything that is lexicographically greater than
"abc"
Example2: =, 'binaryprefix:abc' will match everything whose first 3 characters are
lexicographically equal to "abc"
Example3: !=, 'regexstring:ab*yz' will match everything that doesn't begin with "ab"
and ends with "yz"
Example4: =, 'substring:abc123' will match everything that begins with the substring
"abc123"

Compound Operators
Within an expression, parentheses can be used to group clauses together, and parentheses have the highest
order of precedence.
SKIP and WHILE operators are next, and have the same precedence.

The AND operator is next.


The OR operator is next.

96 | Cloudera Administration

Managing CDH and Managed Services


Examples
A filter string of the form: Filter1 AND Filter2 OR Filter3 will be evaluated as:
(Filter1 AND Filter2) OR Filter3
A filter string of the form: Filter1 AND SKIP Filter2 OR Filter3 will be evaluated
as: (Filter1 AND (SKIP Filter2)) OR Filter3

Filter Types
HBase includes several filter types, as well as the ability to group filters together and create your own custom
filters.
KeyOnlyFilter - takes no arguments. Returns the key portion of each key-value pair.
Syntax: KeyOnlyFilter ()

FirstKeyOnlyFilter - takes no arguments. Returns the key portion of the first key-value pair.
Syntax: FirstKeyOnlyFilter ()

PrefixFilter - takes a single argument, a prefix of a row key. It returns only those key-values present in a row
that start with the specified row prefix
Syntax:

PrefixFilter (<row_prefix>)

Example: PrefixFilter (Row)

ColumnPrefixFilter - takes a single argument, a column prefix. It returns only those key-values present in a
column that starts with the specified column prefix.
Syntax:

ColumnPrefixFilter (<column_prefix>)

Example: ColumnPrefixFilter (Col)

MultipleColumnPrefixFilter - takes a list of column prefixes. It returns key-values that are present in a column
that starts with any of the specified column prefixes.
Syntax: MultipleColumnPrefixFilter (<column_prefix>, <column_prefix>, ,
<column_prefix>)
Example: MultipleColumnPrefixFilter (Col1, Col2)

ColumnCountGetFilter - takes one argument, a limit. It returns the first limit number of columns in the
table.
Syntax:

ColumnCountGetFilter (<limit>)

Example: ColumnCountGetFilter (4)

PageFilter - takes one argument, a page size. It returns page size number of rows from the table.
Syntax:

PageFilter (<page_size>)

Example: PageFilter (2)

Cloudera Administration | 97

Managing CDH and Managed Services


ColumnPaginationFilter - takes two arguments, a limit and offset. It returns limit number of columns after
offset number of columns. It does this for all the rows.
Syntax:

ColumnPaginationFilter (<limit>, <offset>)

Example: ColumnPaginationFilter (3, 5)

InclusiveStopFilter - takes one argument, a row key on which to stop scanning. It returns all key-values
present in rows up to and including the specified row.
Syntax:

InclusiveStopFilter (<stop_row_key>)

Example: InclusiveStopFilter (Row2)

TimeStampsFilter - takes a list of timestamps. It returns those key-values whose timestamps matches any
of the specified timestamps.
Syntax:

TimeStampsFilter (<timestamp>, <timestamp>, ... ,<timestamp>)

Example: TimeStampsFilter (5985489, 48895495, 58489845945)

RowFilter - takes a compare operator and a comparator. It compares each row key with the comparator
using the compare operator and if the comparison returns true, it returns all the key-values in that row.
Syntax:

RowFilter (<compareOp>, <row_comparator>)

Example: RowFilter (<=, binary:xyz)

FamilyFilter - takes a compare operator and a comparator. It compares each family name with the comparator
using the compare operator and if the comparison returns true, it returns all the key-values in that family.
Syntax:

FamilyFilter (<compareOp>, <family_comparator>)

Example: FamilyFilter (>=, binaryprefix:FamilyB)

QualifierFilter - takes a compare operator and a comparator. It compares each qualifier name with the
comparator using the compare operator and if the comparison returns true, it returns all the key-values in
that column.
Syntax:

QualifierFilter (<compareOp>, <qualifier_comparator>)

Example: QualifierFilter (=, substring:Column1)

ValueFilter - takes a compare operator and a comparator. It compares each value with the comparator using
the compare operator and if the comparison returns true, it returns that key-value.
Syntax:

ValueFilter (<compareOp>, <value_comparator>)

Example: ValueFilter (!=, binary:Value)

DependentColumnFilter - takes two arguments required arguments, a family and a qualifier. It tries to locate
this column in each row and returns all key-values in that row that have the same timestamp. If the row
does not contain the specified column, none of the key-values in that row will be returned.
The filter can also take an optional boolean argument, dropDependentColumn. If set to true, the column
used for the filter does not get returned.
The filter can also take two more additional optional arguments, a compare operator and a value comparator,
which are further checks in addition to the family and qualifier. If the dependent column is found, its value

98 | Cloudera Administration

Managing CDH and Managed Services


should also pass the value check. If it does pass the value check, only then is its timestamp taken into
consideration.
Syntax: DependentColumnFilter (<family>, <qualifier>, <boolean>, <compare
operator>, <value comparator)
DependentColumnFilter (<family>, <qualifier>, <boolean>)
DependentColumnFilter (<family>, <qualifier>)
Example: DependentColumnFilter (conf, blacklist, false, >=, zebra)
DependentColumnFilter (conf, blacklist, true)
DependentColumnFilter (conf, blacklist)

SingleColumnValueFilter - takes a column family, a qualifier, a compare operator and a comparator. If the
specified column is not found, all the columns of that row will be emitted. If the column is found and the
comparison with the comparator returns true, all the columns of the row will be emitted. If the condition
fails, the row will not be emitted.
This filter also takes two additional optional boolean arguments, filterIfColumnMissing and
setLatestVersionOnly.
If the filterIfColumnMissing flag is set to true, the columns of the row will not be emitted if the specified
column to check is not found in the row. The default value is false.
If the setLatestVersionOnly flag is set to false, it will test previous versions (timestamps) in addition to
the most recent. The default value is true.
These flags are optional and dependent on each other. You must set neither or both of them together.
Syntax: SingleColumnValueFilter (<family>, <qualifier>, <compare operator>,
<comparator>, <filterIfColumnMissing_boolean>, <latest_version_boolean>)
Syntax: SingleColumnValueFilter (<family>, <qualifier>, <compare operator>,
<comparator>)
Example: SingleColumnValueFilter (FamilyA, Column1, <=, abc, true, false)
Example: SingleColumnValueFilter ('FamilyA, Column1, <=, abc)

SingleColumnValueExcludeFilter - takes the same arguments and behaves same as


SingleColumnValueFilter. However, if the column is found and the condition passes, all the columns of
the row will be emitted except for the tested column value.
Syntax: SingleColumnValueExcludeFilter (<family>, <qualifier>, <compare operators>,
<comparator>, <latest_version_boolean>, <filterIfColumnMissing_boolean>)
Syntax: SingleColumnValueExcludeFilter (<family>, <qualifier>, <compare operator>
<comparator>)
Example: SingleColumnValueExcludeFilter (FamilyA, Column1, <=, abc, false,
true)
Example: SingleColumnValueExcludeFilter (FamilyA, Column1, <=, abc)

ColumnRangeFilter - takes either minColumn, maxColumn, or both. Returns only those keys with columns
that are between minColumn and maxColumn. It also takes two boolean variables to indicate whether to
include the minColumn and maxColumn or not. If you dont want to set the minColumn or the maxColumn, you
can pass in an empty argument.
Syntax: ColumnRangeFilter (<minColumn >, <minColumnInclusive_bool>,
<maxColumn>, <maxColumnInclusive_bool>)
Example: ColumnRangeFilter (abc, true, xyz, false)

Custom Filter - You can create a custom filter by implementing the Filter class. The JAR must be available
on all region servers.

Cloudera Administration | 99

Managing CDH and Managed Services


HBase Shell Example
This example scans the 'users' table for rows where the contents of the cf:name column equals the string 'abc'.
hbase> scan 'users', { FILTER => SingleColumnValueFilter.new(Bytes.toBytes('cf'),
Bytes.toBytes('name'), CompareFilter::CompareOp.valueOf('EQUAL'),
BinaryComparator.new(Bytes.toBytes('abc')))}

Java API Example


This example, taken from the HBase unit test found in
hbase-server/src/test/java/org/apache/hadoop/hbase/filter/TestSingleColumnValueFilter.java

, shows how to use the Java API to implement several different filters..
/**
*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
*
http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hbase.filter;
import static org.junit.Assert.assertFalse;
import static org.junit.Assert.assertTrue;
import java.util.regex.Pattern;
import
import
import
import
import
import
import

org.apache.hadoop.hbase.KeyValue;
org.apache.hadoop.hbase.SmallTests;
org.apache.hadoop.hbase.filter.CompareFilter.CompareOp;
org.apache.hadoop.hbase.util.Bytes;
org.junit.Before;
org.junit.Test;
org.junit.experimental.categories.Category;

/**
* Tests the value filter
*/
@Category(SmallTests.class)
public class TestSingleColumnValueFilter {
private static final byte[] ROW = Bytes.toBytes("test");
private static final byte[] COLUMN_FAMILY = Bytes.toBytes("test");
private static final byte [] COLUMN_QUALIFIER = Bytes.toBytes("foo");
private static final byte[] VAL_1 = Bytes.toBytes("a");
private static final byte[] VAL_2 = Bytes.toBytes("ab");
private static final byte[] VAL_3 = Bytes.toBytes("abc");
private static final byte[] VAL_4 = Bytes.toBytes("abcd");
private static final byte[] FULLSTRING_1 =
Bytes.toBytes("The quick brown fox jumps over the lazy dog.");
private static final byte[] FULLSTRING_2 =
Bytes.toBytes("The slow grey fox trips over the lazy dog.");
private static final String QUICK_SUBSTR = "quick";
private static final String QUICK_REGEX = ".+quick.+";
private static final Pattern QUICK_PATTERN = Pattern.compile("QuIcK",
Pattern.CASE_INSENSITIVE | Pattern.DOTALL);
Filter basicFilter;
Filter nullFilter;
Filter substrFilter;

100 | Cloudera Administration

Managing CDH and Managed Services


Filter regexFilter;
Filter regexPatternFilter;
@Before
public void setUp() throws Exception {
basicFilter = basicFilterNew();
nullFilter = nullFilterNew();
substrFilter = substrFilterNew();
regexFilter = regexFilterNew();
regexPatternFilter = regexFilterNew(QUICK_PATTERN);
}
private Filter basicFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.GREATER_OR_EQUAL, VAL_2);
}
private Filter nullFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.NOT_EQUAL,
new NullComparator());
}
private Filter substrFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.EQUAL,
new SubstringComparator(QUICK_SUBSTR));
}
private Filter regexFilterNew() {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.EQUAL,
new RegexStringComparator(QUICK_REGEX));
}
private Filter regexFilterNew(Pattern pattern) {
return new SingleColumnValueFilter(COLUMN_FAMILY, COLUMN_QUALIFIER,
CompareOp.EQUAL,
new RegexStringComparator(pattern.pattern(), pattern.flags()));
}
private void basicFilterTests(SingleColumnValueFilter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_2);
assertTrue("basicFilter1", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_3);
assertTrue("basicFilter2", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_4);
assertTrue("basicFilter3", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("basicFilterNotNull", filter.filterRow());
filter.reset();
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_1);
assertTrue("basicFilter4", filter.filterKeyValue(kv) == Filter.ReturnCode.NEXT_ROW);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_2);
assertTrue("basicFilter4", filter.filterKeyValue(kv) == Filter.ReturnCode.NEXT_ROW);
assertFalse("basicFilterAllRemaining", filter.filterAllRemaining());
assertTrue("basicFilterNotNull", filter.filterRow());
filter.reset();
filter.setLatestVersionOnly(false);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_1);
assertTrue("basicFilter5", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, VAL_2);
assertTrue("basicFilter5", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("basicFilterNotNull", filter.filterRow());
}

Cloudera Administration | 101

Managing CDH and Managed Services


private void nullFilterTests(Filter filter) throws Exception {
((SingleColumnValueFilter) filter).setFilterIfMissing(true);
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER, FULLSTRING_1);
assertTrue("null1", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("null1FilterRow", filter.filterRow());
filter.reset();
kv = new KeyValue(ROW, COLUMN_FAMILY, Bytes.toBytes("qual2"), FULLSTRING_2);
assertTrue("null2", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertTrue("null2FilterRow", filter.filterRow());
}
private void substrFilterTests(Filter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_1);
assertTrue("substrTrue",
filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_2);
assertTrue("substrFalse", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("substrFilterAllRemaining", filter.filterAllRemaining());
assertFalse("substrFilterNotNull", filter.filterRow());
}
private void regexFilterTests(Filter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_1);
assertTrue("regexTrue",
filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_2);
assertTrue("regexFalse", filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("regexFilterAllRemaining", filter.filterAllRemaining());
assertFalse("regexFilterNotNull", filter.filterRow());
}
private void regexPatternFilterTests(Filter filter)
throws Exception {
KeyValue kv = new KeyValue(ROW, COLUMN_FAMILY, COLUMN_QUALIFIER,
FULLSTRING_1);
assertTrue("regexTrue",
filter.filterKeyValue(kv) == Filter.ReturnCode.INCLUDE);
assertFalse("regexFilterAllRemaining", filter.filterAllRemaining());
assertFalse("regexFilterNotNull", filter.filterRow());
}
private Filter serializationTest(Filter filter)
throws Exception {
// Decompose filter to bytes.
byte[] buffer = filter.toByteArray();
// Recompose filter.
Filter newFilter = SingleColumnValueFilter.parseFrom(buffer);
return newFilter;
}
/**
* Tests identification of the stop row
* @throws Exception
*/
@Test
public void testStop() throws Exception {
basicFilterTests((SingleColumnValueFilter) basicFilter);
nullFilterTests(nullFilter);
substrFilterTests(substrFilter);
regexFilterTests(regexFilter);
regexPatternFilterTests(regexPatternFilter);
}
/**
* Tests serialization

102 | Cloudera Administration

Managing CDH and Managed Services


* @throws Exception
*/
@Test
public void testSerialization() throws Exception {
Filter newFilter = serializationTest(basicFilter);
basicFilterTests((SingleColumnValueFilter)newFilter);
newFilter = serializationTest(nullFilter);
nullFilterTests(newFilter);
newFilter = serializationTest(substrFilter);
substrFilterTests(newFilter);
newFilter = serializationTest(regexFilter);
regexFilterTests(newFilter);
newFilter = serializationTest(regexPatternFilter);
regexPatternFilterTests(newFilter);
}
}

Writing Data to HBase


To write data to HBase, you use methods of the HTableInterface class. You can use the Java API directly, or
use HBase Shell, Thrift API, REST API, or another client which uses the Java API indirectly. When you issue a Put,
the coordinates of the data are the row, the column, and the timestamp. The timestamp is unique per version
of the cell, and can be generated automatically or specified programmatically by your application, and must be
a long integer.
Variations on Put
There are several different ways to write data into HBase. Some of them are listed below.
A Put operation writes data into HBase.
A Delete operation deletes data from HBase. What actually happens during a Delete depends upon several
factors.
A CheckAndPut operation performs a Scan before attempting the Put, and only does the Put if a value
matches what is expected, and provides row-level atomicity.
A CheckAndDelete operation performs a Scan before attempting the Delete, and only does the Delete if
a value matches what is expected.
An Increment operation increments values of one or more columns within a single row, and provides row-level
atomicity.
Refer to the API documentation for a full list of methods provided for writing data to HBase.Different methods
require different access levels and have other differences.
Versions
When you put data into HBase, a timestamp is required. The timestamp can be generated automatically by the
RegionServer or can be supplied by you. The timestamp must be unique per version of a given cell, because the
timestamp identifies the version. To modify a previous version of a cell, for instance, you would issue a Put with
a different value for the data itself, but the same timestamp.
HBase's behavior regarding versions is highly configurable. The maximum number of versions defaults to 1 in
CDH 5, and 3 in previous versions. You can change the default value for HBase by configuring
hbase.column.max.version in hbase-site.xml, either via an advanced configuration snippet if you use
Cloudera Manager, or by editing the file directly otherwise.
You can also configure the maximum and minimum number of versions to keep for a given column, or specify
a default time-to-live (TTL), which is the number of seconds before a version is deleted. The following examples
all use alter statements in HBase Shell to create new column families with the given characteristics, but you

Cloudera Administration | 103

Managing CDH and Managed Services


can use the same syntax when creating a new table or to alter an existing column family. This is only a fraction
of the options you can specify for a given column family.
hbase> alter t1, NAME => f1, VERSIONS => 5
hbase> alter t1, NAME => f1, MIN_VERSIONS => 2
hbase> alter t1, NAME => f1, TTL => 15

HBase sorts the versions of a cell from newest to oldest, by sorting the timestamps lexicographically. When a
version needs to be deleted because a threshold has been reached, HBase always chooses the "oldest" version,
even if it is in fact the most recent version to be inserted. Keep this in mind when designing your timestamps.
Consider using the default generated timestamps and storing other version-specific data elsewhere in the row,
such as in the row key. If MIN_VERSIONS and TTL conflict, MIN_VERSIONS takes precedence.
Deletion
When you request for HBase to delete data, either explicitly via a Delete method or implicitly via a threshold
such as the maximum number of versions or the TTL, HBase does not delete the data immediately. Instead, it
writes a deletion marker, called a tombstone, to the HFile, which is the physical file where a given RegionServer
stores its region of a column family. The tombstone markers are processed during major compaction operations,
when HFiles are rewritten without the deleted data included.
Even after major compactions, "deleted" data may not actually be deleted. You can specify the
KEEP_DELETED_CELLS option for a given column family, and the tombstones will be preserved in the HFile even
after major compaction. One scenario where this approach might be useful is for data retention policies.
Another reason deleted data may not actually be deleted is if the data would be required to restore a table from
a snapshot which has not been deleted. In this case, the data is moved to an archive during a major compaction,
and only deleted when the snapshot is deleted. This is a good reason to monitor the number of snapshots saved
in HBase.
Examples
This abbreviated example writes data to an HBase table using HBase Shell and then scans the table to show
the result.
hbase> put 'test', 'row1', 'cf:a', 'value1'
0 row(s) in 0.1770 seconds
hbase> put 'test', 'row2', 'cf:b', 'value2'
0 row(s) in 0.0160 seconds
hbase> put 'test', 'row3', 'cf:c',
0 row(s) in 0.0260 seconds
hbase> scan 'test'
ROW
COLUMN+CELL
row1
column=cf:a,
row2
column=cf:b,
row3
column=cf:c,
3 row(s) in 0.0440 seconds

'value3'

timestamp=1403759475114, value=value1
timestamp=1403759492807, value=value2
timestamp=1403759503155, value=value3

This abbreviated example uses the HBase API to write data to an HBase table, using the automatic timestamp
created by the Region Server.
publicstaticfinalbyte[] CF = "cf".getBytes();
publicstaticfinalbyte[] ATTR = "attr".getBytes();
...
Put put = new Put(Bytes.toBytes(row));
put.add(CF, ATTR, Bytes.toBytes( data));
htable.put(put);

104 | Cloudera Administration

Managing CDH and Managed Services


This example uses the HBase API to write data to an HBase table, specifying the timestamp.
publicstaticfinalbyte[] CF = "cf".getBytes();
publicstaticfinalbyte[] ATTR = "attr".getBytes();
...
Put put = new Put( Bytes.toBytes(row));
long explicitTimeInMs = 555; // just an example
put.add(CF, ATTR, explicitTimeInMs, Bytes.toBytes(data));
htable.put(put);

Further Reading
Refer to the HTableInterface and HColumnDescriptor API documentation for more details about configuring
tables and columns, as well as reading and writing to HBase.
Refer to the Apache HBase Reference Guide for more in-depth information about HBase, including details
about versions and deletions not covered here.

Importing Data Into HBase


The method you use for importing data into HBase depends on several factors:

The location, size, and format of your existing data


Whether you need to import data once or periodically over time
Whether you want to import the data in bulk or stream it into HBase regularly
How fresh the HBase data needs to be

This topic helps you choose the correct method or composite of methods and provides example workflows for
each method.
Always run HBase administrative commands as the HBase user (typically hbase).
Choosing the Right Import Method
If the data is already in an HBase table:
To move the data from one HBase cluster to another, use snapshot and either the clone_snapshot or
ExportSnapshot utility; or, use the CopyTable utility.
To move the data from one HBase cluster to another without downtime on either cluster, use replication.
To migrate data between HBase version that are not wire compatible, such as from CDH 4 to CDH 5, see
Importing HBase Data From CDH 4 to CDH 5 on page 106.
If the data currently exists outside HBase:
If possible, write the data to HFile format, and use a BulkLoad to import it into HBase. The data is immediately
available to HBase and you can bypass the normal write path, increasing efficiency.
If you prefer not to use bulk loads, and you are using a tool such as Pig, you can use it to import your data.
If you need to stream live data to HBase instead of import in bulk:
Write a Java client using the Java API, or use the Apache Thrift Proxy API to write a client in a language
supported by Thrift.
Stream data directly into HBase using the REST Proxy API in conjunction with an HTTP client such as wget
or curl.
Use Flume or Spark.
Most likely, at least one of these methods works in your situation. If not, you can use MapReduce directly. Test
the most feasible methods with a subset of your data to determine which one is optimal.

Cloudera Administration | 105

Managing CDH and Managed Services


Using CopyTable
CopyTable uses HBase read and write paths to copy part or all of a table to a new table in either the same
cluster or a different cluster. CopyTable causes read load when reading from the source, and write load when

writing to the destination. Region splits occur on the destination table in real time as needed. To avoid these
issues, use snapshot and export commands instead of CopyTable. Alternatively, you can pre-split the
destination table to avoid excessive splits. The destination table can be partitioned differently from the source
table. See this section of the Apache HBase documentation for more information.
Edits to the source table after the CopyTable starts are not copied, so you may need to do an additional
CopyTable operation to copy new data into the destination table. Run CopyTable as follows, using --help to
see details about possible parameters.
$ ./bin/hbase org.apache.hadoop.hbase.mapreduce.CopyTable --help
Usage: CopyTable [general options] [--starttime=X] [--endtime=Y] [--new.name=NEW]
[--peer.adr=ADR] <tablename>

The starttime/endtime and startrow/endrow pairs function in a similar way: if you leave out the first of the
pair, the first timestamp or row in the table is the starting point. Similarly, if you leave out the second of the
pair, the operation continues until the end of the table. To copy the table to a new table in the same cluster, you
must specify --new.name, unless you want to write the copy back to the same table, which would add a new
version of each cell (with the same data), or just overwrite the cell with the same value if the maximum number
of versions is set to 1 (the default in CDH 5). To copy the table to a new table in a different cluster, specify
--peer.adr and optionally, specify a new table name.
The following example creates a new table using HBase Shell in non-interactive mode, and then copies data in
two ColumnFamilies in rows starting with timestamp 1265875194289 and including the last row before the
CopyTable started, to the new table.
$ echo create 'NewTestTable', 'cf1', 'cf2', 'cf3' | bin/hbase shell --non-interactive
$ bin/hbase org.apache.hadoop.hbase.mapreduce.CopyTable --starttime=1265875194289
--families=cf1,cf2,cf3 --new.name=NewTestTable TestTable

In CDH 5, snapshots are recommended instead of CopyTable for most situations.


Importing HBase Data From CDH 4 to CDH 5
CDH 4 and CDH 5 are not wire-compatible, so import methods such as CopyTable will not work. Instead, you can
use separate export and import operations using distcp, or you can copy the table's HFiles using HDFS utilities
and upgrade the HFiles in place. The first option is preferred unless the size of the table is too large to be practical
and the export or import will take too long. The import/export mechanism gives you flexibility and allows you
to run exports as often as you need, for an ongoing period of time. This would allow you to test CDH 5 with your
production data before finalizing your upgrade, for instance.
Import and Export Data Using DistCP
1. On the CDH 4 cluster, export the contents of the table to sequence files in a given directory using a command
like the following.
$ sudo -u hdfs hbase org.apache.hadoop.hbase.mapreduce.Export <tablename>
/export_directory

The sequence files are located in the /export_directory directory.


2. Copy the contents of /export_directory to the CDH 5 cluster using distcp or through a filesystem
accessible from nodes on both clusters. If you use distcp, the following is an example command.
$ sudo -u hdfs hadoop distcp -p -update -skipcrccheck
hftp://cdh4-namenode:port/export_directory hdfs://cdh5-namenode/import_directory

106 | Cloudera Administration

Managing CDH and Managed Services


3. Create the table on the CDH 5 cluster using HBase Shell. Column families must be identical to the table on
the CDH 4 cluster.
4. Import the sequence file into the newly-created table.
$ sudo -u hdfs hbase -Dhbase.import.version=0.94
org.apache.hadoop.hbase.mapreduce.Import t1 /import_directory

Copy and Upgrade the HFiles


If exporting and importing the data is not feasible because of the size of the data or other reasons, or you know
that the import will be a one-time occurrence, you can copy the HFiles directly from the CDH 4 cluster's HDFS
filesystem to the CDH 5 cluster's HDFS filesystem, and upgrade the HFiles in place.
Warning: Only use this procedure if the destination cluster is a brand new HBase cluster with empty
tables, and is not currently hosting any data. If this is not the case, or if you are unsure, contact
Cloudera Support before following this procedure.
1. Use the distcp command on the CDH 5 cluster to copy the HFiles from the CDH 4 cluster.
$ sudo -u hdfs hadoop distcp -p -update -skipcrccheck
webhdfs://cdh4-namenode:http-port/hbase hdfs://cdh5-namenode:rpc-port/hbase

2. In the destination cluster, upgrade the HBase tables. In Cloudera Manager, go to Cluster > HBase and choose
Upgrade HBase from the Action menu. This checks that the HBase tables can be upgraded, and then upgrades
them.
3. Start HBase on the CDH 5 cluster. The upgraded tables are available. Verify the data and confirm that no
errors are logged.
Using Snapshots
As of CDH 4.7, Cloudera recommends snapshots instead of CopyTable where possible. A snapshot captures the
state of a table at the time the snapshot was taken. Because no data is copied when a snapshot is taken, the
process is very quick. As long as the snapshot exists, cells in the snapshot are never deleted from HBase, even
if they are explicitly deleted by the API. Instead, they are archived so that the snapshot can restore the table to
its state at the time of the snapshot.
After taking a snapshot, use the clone_snapshot command to copy the data to a new (immediately enabled)
table in the same cluster, or the Export utility to create a new table based on the snapshot, in the same cluster
or a new cluster. This is a copy-on-write operation. The new table shares HFiles with the original table until
writes occur in the new table but not the old table, or until a compaction or split occurs in either of the tables.
This can improve performance in the short term compared to CopyTable.
To export the snapshot to a new cluster, use the ExportSnapshot utility, which uses MapReduce to copy the
snapshot to the new cluster. Run the ExportSnapshot utility on the source cluster, as a user with HBase and
HDFS write permission on the destination cluster, and HDFS read permission on the source cluster. This creates
the expected amount of IO load on the destination cluster. Optionally, you can limit bandwidth consumption,
which affects IO on the destination cluster. After the ExportSnapshot operation completes, you can see the
snapshot in the new cluster using the list_snapshot command, and you can use the clone_snapshot
command to create the table in the new cluster from the snapshot.
For full instructions for the snapshot and clone_snapshot HBase Shell commands, run the HBase Shell and
type help snapshot. The following example takes a snapshot of a table, uses it to clone the table to a new
table in the same cluster, and then uses the ExportSnapshot utility to copy the table to a different cluster, with
16 mappers and limited to 200 Mb/sec bandwidth.
$ bin/hbase shell
hbase(main):005:0> snapshot 'TestTable', 'TestTableSnapshot'
0 row(s) in 2.3290 seconds

Cloudera Administration | 107

Managing CDH and Managed Services


hbase(main):006:0> clone_snapshot 'TestTableSnapshot', 'NewTestTable'
0 row(s) in 1.3270 seconds
hbase(main):007:0> describe 'NewTestTable'
DESCRIPTION
ENABLED
'NewTestTable', {NAME => 'cf1', DATA_BLOCK_ENCODING true
=> 'NONE', BLOOMFILTER => 'ROW', REPLICATION_SCOPE
=> '0', VERSIONS => '1', COMPRESSION => 'NONE', MI
N_VERSIONS => '0', TTL => 'FOREVER', KEEP_DELETED_C
ELLS => 'false', BLOCKSIZE => '65536', IN_MEMORY =>
'false', BLOCKCACHE => 'true'}, {NAME => 'cf2', DA
TA_BLOCK_ENCODING => 'NONE', BLOOMFILTER => 'ROW',
REPLICATION_SCOPE => '0', VERSIONS => '1', COMPRESS
ION => 'NONE', MIN_VERSIONS => '0', TTL => 'FOREVER
', KEEP_DELETED_CELLS => 'false', BLOCKSIZE => '655
36', IN_MEMORY => 'false', BLOCKCACHE => 'true'}
1 row(s) in 0.1280 seconds
hbase(main):008:0> quit
$ hbase org.apache.hadoop.hbase.snapshot.ExportSnapshot -snapshot TestTableSnapshot
-copy-to file:///tmp/hbase -mappers 16 -bandwidth 200
14/10/28 21:48:16 INFO snapshot.ExportSnapshot: Copy Snapshot Manifest
14/10/28 21:48:17 INFO client.RMProxy: Connecting to ResourceManager at
a1221.halxg.cloudera.com/10.20.188.121:8032
14/10/28 21:48:19 INFO snapshot.ExportSnapshot: Loading Snapshot 'TestTableSnapshot'
hfile list
14/10/28 21:48:19 INFO Configuration.deprecation: hadoop.native.lib is deprecated.
Instead, use io.native.lib.available
14/10/28 21:48:19 INFO util.FSVisitor: No logs under
directory:hdfs://a1221.halxg.cloudera.com:8020/hbase/.hbase-snapshot/TestTableSnapshot/WALs
14/10/28 21:48:20 INFO mapreduce.JobSubmitter: number of splits:0
14/10/28 21:48:20 INFO mapreduce.JobSubmitter: Submitting tokens for job:
job_1414556809048_0001
14/10/28 21:48:20 INFO impl.YarnClientImpl: Submitted application
application_1414556809048_0001
14/10/28 21:48:20 INFO mapreduce.Job: The url to track the job:
http://a1221.halxg.cloudera.com:8088/proxy/application_1414556809048_0001/
14/10/28 21:48:20 INFO mapreduce.Job: Running job: job_1414556809048_0001
14/10/28 21:48:36 INFO mapreduce.Job: Job job_1414556809048_0001 running in uber mode
: false
14/10/28 21:48:36 INFO mapreduce.Job: map 0% reduce 0%
14/10/28 21:48:37 INFO mapreduce.Job: Job job_1414556809048_0001 completed successfully
14/10/28 21:48:37 INFO mapreduce.Job: Counters: 2
Job Counters
Total time spent by all maps in occupied slots (ms)=0
Total time spent by all reduces in occupied slots (ms)=0
14/10/28 21:48:37 INFO snapshot.ExportSnapshot: Finalize the Snapshot Export
14/10/28 21:48:37 INFO snapshot.ExportSnapshot: Verify snapshot integrity
14/10/28 21:48:37 INFO Configuration.deprecation: fs.default.name is deprecated. Instead,
use fs.defaultFS
14/10/28 21:48:37 INFO snapshot.ExportSnapshot: Export Completed: TestTableSnapshot

The bold italic line contains the URL from which you can track the ExportSnapshot job. When it finishes, a new
set of HFiles, comprising all of the HFiles that were part of the table when the snapshot was taken, is created
at the HDFS location you specified.
You can use the SnapshotInfo command-line utility included with HBase to verify or debug snapshots.
Using BulkLoad
HBase uses the well-known HFile format to store its data on disk. In many situations, writing HFiles
programmatically with your data, and bulk-loading that data into HBase on the RegionServer, has advantages
over other data ingest mechanisms. BulkLoad operations bypass the write path completely, providing the
following benefits:
The data is available to HBase immediately but does cause additional load or latency on the cluster when it
appears.
BulkLoad operations do not use the write-ahead log (WAL) and do not cause flushes or split storms.
BulkLoad operations do not cause excessive garbage collection.

108 | Cloudera Administration

Managing CDH and Managed Services


Note: Because they bypass the WAL, BulkLoad operations are not propagated between clusters
using replication. If you need the data on all replicated clusters, you must perform the BulkLoad
on each cluster.
If you use BulkLoads with HBase, your workflow is similar to the following:
1. Extract your data from its existing source. For instance, if your data is in a MySQL database, you might run
the mysqldump command. The process you use depends on your data. If your data is already in TSV or CSV
format, skip this step and use the included ImportTsv utility to process your data into HFiles. See the
ImportTsv documentation for details.
2. Process your data into HFile format. See http://hbase.apache.org/book/hfile_format.html for details about
HFile format. Usually you use a MapReduce job for the conversion, and you often need to write the Mapper
yourself because your data is unique. The job must to emit the row key as the Key, and either a KeyValue,
a Put, or a Delete as the Value. The Reducer is handled by HBase; configure it using
HFileOutputFormat.configureIncrementalLoad() and it does the following:

Inspects the table to configure a total order partitioner


Uploads the partitions file to the cluster and adds it to the DistributedCache
Sets the number of reduce tasks to match the current number of regions
Sets the output key/value class to match HFileOutputFormat requirements
Sets the Reducer to perform the appropriate sorting (either KeyValueSortReducer or PutSortReducer)

3. One HFile is created per region in the output folder. Input data is almost completely re-written, so you need
available disk space at least twice the size of the original data set. For example, for a 100 GB output from
mysqldump, you should have at least 200 GB of available disk space in HDFS. You can delete the original input
file at the end of the process.
4. Load the files into HBase. Use the LoadIncrementalHFiles command (more commonly known as the
completebulkload tool), passing it a URL that locates the files in HDFS. Each file is loaded into the relevant
region on the RegionServer for the region. You can limit the number of versions that are loaded by passing
the --versions= N option, where N is the maximum number of versions to include, from newest to
oldest (largest timestamp to smallest timestamp).
If a region was split after the files were created, the tool automatically splits the HFile according to the new
boundaries. This process is inefficient, so if your table is being written to by other processes, you should load
as soon as the transform step is done.
The following illustration shows the full BulkLoad process.

Use Cases for BulkLoad:


Loading your original dataset into HBase for the first time - Your initial dataset might be quite large, and
bypassing the HBase write path can speed up the process considerably.
Cloudera Administration | 109

Managing CDH and Managed Services


Incremental Load - To load new data periodically, use BulkLoad to import it in batches at your preferred
intervals. This alleviates latency problems and helps you to achieve service-level agreements (SLAs). However,
one trigger for compaction is the number of HFiles on a RegionServer. Therefore, importing a large number
of HFiles at frequent intervals can cause major compactions to happen more often than they otherwise
would, negatively impacting performance. You can mitigate this by tuning the compaction settings such that
the maximum number of HFiles that can be present without triggering a compaction is very high, and relying
on other factors, such as the size of the Memstore, to trigger compactions.
Data needs to originate elsewhere - If an existing system is capturing the data you want to have in HBase
and needs to remain active for business reasons, you can periodically BulkLoad data from the system into
HBase so that you can perform operations on it without impacting the system.
For more information and examples, as well as an explanation of the ImportTsv utility, which can be used to
import data in text-delimited formats such as CSV, see this post on the Cloudera Blog.
Using Cluster Replication
If your data is already in an HBase cluster, replication is useful for getting the data into additional HBase clusters.
In HBase, cluster replication refers to keeping one cluster state synchronized with that of another cluster, using
the write-ahead log (WAL) of the source cluster to propagate the changes. Replication is enabled at column
family granularity. Before enabling replication for a column family, create the table and all column families to
be replicated, on the destination cluster.
Cluster replication uses an active-push methodology. An HBase cluster can be a source (also called active,
meaning that it is the originator of new data), a destination (also called passive, meaning that it receives data
via replication), or can fulfill both roles at once. Replication is asynchronous, and the goal of replication is
consistency.
When data is replicated from one cluster to another, the original source of the data is tracked with a cluster ID,
which is part of the metadata. In CDH 5, all clusters that have already consumed the data are also tracked. This
prevents replication loops.
Common Replication Topologies
A central source cluster might propagate changes to multiple destination clusters, for failover or due to
geographic distribution.
A source cluster might push changes to a destination cluster, which might also push its own changes back
to the original cluster.
Many different low-latency clusters might push changes to one centralized cluster for backup or
resource-intensive data-analytics jobs. The processed data might then be replicated back to the low-latency
clusters.
Multiple levels of replication can be chained together to suit your needs. The following diagram shows a
hypothetical scenario. Use the arrows to follow the data paths.

110 | Cloudera Administration

Managing CDH and Managed Services

At the top of the diagram, the San Jose and Tokyo clusters, shown in red, replicate changes to each other, and
each also replicates changes to a User Data and a Payment Data cluster.
Each cluster in the second row, shown in blue, replicates its changes to the All Data Backup 1 cluster, shown
in grey. The All Data Backup 1 cluster replicates changes to the All Data Backup 2 cluster (also shown in
grey), as well as the Data Analysis cluster (shown in green). All Data Backup 2 also propagates any of its
own changes back to All Data Backup 1.
The Data Analysis cluster runs MapReduce jobs on its data, and then pushes the processed data back to the
San Jose and Tokyo clusters.
Configuring Clusters for Replication
To configure your clusters for replication, see HBase Replication on page 405 and Configuring Secure HBase
Replication. The following is a high-level overview of the steps to enable replication.
1. Configure and start the source and destination clusters. Create tables with the same names and column
families on both the source and destination clusters, so that the destination cluster knows where to store
data it receives. All hosts in the source and destination clusters should be reachable to each other.
2. On the source cluster, enable replication in Cloudera Manager, or by setting hbase.replication to true in
hbase-site.xml.
3. On the source cluster, in HBase Shell, add the destination cluster as a peer, using the add_peer command.
The syntax is as follows:
add_peer 'ID', 'CLUSTER_KEY'

The ID must be a short integer. To compose the CLUSTER_KEY, use the following template:
hbase.zookeeper.quorum:hbase.zookeeper.property.clientPort:zookeeper.znode.parent

If both clusters use the same ZooKeeper cluster, you must use a different zookeeper.znode.parent, because
they cannot write in the same folder.

Cloudera Administration | 111

Managing CDH and Managed Services


4. On the source cluster, configure each column family to be replicated by setting its REPLICATION_SCOPE to
1, using commands such as the following in HBase Shell.
hbase> disable 'example_table'
hbase> alter 'example_table', {NAME => 'example_family', REPLICATION_SCOPE => '1'}
hbase> enable 'example_table'

5. Verify that replication is occurring by examining the logs on the source cluster for messages such as the
following.
Considering 1 rs, with ratio 0.1
Getting 1 rs from peer cluster # 0
Choosing peer 10.10.1.49:62020

6. To verify the validity of replicated data, use the included VerifyReplication MapReduce job on the source
cluster, providing it with the ID of the replication peer and table name to verify. Other options are available,
such as a time range or specific families to verify.
The command has the following form:
hbase org.apache.hadoop.hbase.mapreduce.replication.VerifyReplication
[--starttime=timestamp1] [--stoptime=timestamp] [--families=comma separated list
of families] <peerId> <tablename>

The VerifyReplication command prints GOODROWS and BADROWS counters to indicate rows that did and
did not replicate correctly.
Note:
Some changes are not replicated and must be propagated by other means, such as Snapshots or
CopyTable. See Initiating Replication When Data Already Exists on page 409 for more details.
Data that existed in the master before replication was enabled.
Operations that bypass the WAL, such as when using BulkLoad or API calls such as
writeToWal(false).
Table schema modifications.

Using Pig and HCatalog


Apache Pig is a platform for analyzing large data sets using a high-level language. Apache HCatalog is a
sub-project of Apache Hive, which enables reading and writing of data from one Hadoop utility to another. You
can use a combination of Pig and HCatalog to import data into HBase. The initial format of your data and other
details about your infrastructure determine the steps you follow to accomplish this task. The following simple
example assumes that you can get your data into a TSV (text-separated value) format, such as a tab-delimited
or comma-delimited text file.
1. Format the data as a TSV file. You can work with other file formats; see the Pig and HCatalog project
documentation for more details.
The following example shows a subset of data from Google's NGram Dataset, which shows the frequency of
specific phrases or letter-groupings found in publications indexed by Google. Here, the first column has been
added to this dataset as the row ID. It is formulated by combining the n-gram itself (in this case, Zones) with
the line number of the file in which it is found (z_LINE_NUM). This creates a format such as "Zones_z_6230867."
The second column is the n-gram itself, the third column is the year of occurrence, the fourth column is the
frequency of occurrence of that Ngram in that year, and the fifth column is the number of distinct publications.
This extract is from the z file of the 1-gram dataset from version 20120701. The data is truncated at the ...

112 | Cloudera Administration

Managing CDH and Managed Services


mark, for the sake of readability of this document. In most real-world scenarios, you will not work with tables
that have five columns. Most HBase tables have one or two columns.
Zones_z_6230867
Zones_z_6230868
Zones_z_6230869
Zones_z_6230870
...
Zones_z_6231150
Zones_z_6231151
Zones_z_6231152
Zones_z_6231153
Zones_z_6231154
Zones_z_6231155
Zones_z_6231156
Zones_z_6231157
Zones_z_6231158
Zones_z_6231159
Zones_z_6231160
Zones_z_6231161
Zones_z_6231162

Zones
Zones
Zones
Zones

1507
1638
1656
1681

1
1
2
8

1
1
1
2

Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones
Zones

1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008

17868
21296
20365
20288
22996
20469
21338
29724
23334
24300
22362
22101
21037

4356
4675
4972
5021
5714
5470
5946
6446
6524
6580
6707
6798
6328

2. Using the hadoop fs command, put the data into HDFS. This example places the file into an /imported_data/
directory.
$ hadoop fs -put zones_frequency.tsv /imported_data/

3. Create and register a new HBase table in HCatalog, using the hcat command, passing it a DDL file to represent
your table. You could also register an existing HBase table, using the same command. The DDL file format
is specified as part of the Hive REST API. The following example illustrates the basic mechanism.
CREATE TABLE
zones_frequency_table (id STRING, ngram STRING, year STRING, freq STRING, sources
STRING)
STORED BY 'org.apache.hcatalog.hbase.HBaseHCatStorageHandler'
TBLPROPERTIES (
'hbase.table.name' = 'zones_frequency_table',
'hbase.columns.mapping' = 'd:ngram,d:year,d:freq,d:sources',
'hcat.hbase.output.bulkMode' = 'true'
);
$ hcat -f zones_frequency_table.ddl

4. Create a Pig file to process the TSV file created in step 1, using the DDL file created in step 3. Modify the file
names and other parameters in this command to match your values if you use data different from this
working example. USING PigStorage('\t') indicates that the input file is tab-delimited. For more details
about Pig syntax, see the Pig Latin reference documentation.
A = LOAD 'hdfs:///imported_data/zones_frequency.tsv' USING PigStorage('\t') AS
(id:chararray, ngram:chararray, year:chararray, freq:chararray, sources:chararray);
-- DUMP A;
STORE A INTO 'zones_frequency_table' USING org.apache.hcatalog.pig.HCatStorer();

Save the file as zones.bulkload.pig.


5. Use the pig command to bulk-load the data into HBase.
$ pig -useHCatalog zones.bulkload.pig

The data is now in HBase and is available to use.

Cloudera Administration | 113

Managing CDH and Managed Services


Using the Java API
The Java API is the most common mechanism for getting data into HBase, through Put operations. The Thrift
and REST APIs, as well as the HBase Shell, use the Java API. The following simple example ouses the Java API
to put data into an HBase table. The Java API traverses the entire write path and can cause compactions and
region splits, which can adversely affect performance.
...
HTable table = null;
try {
table = myCode.createTable(tableName, fam);
int i = 1;
List<Put> puts = new ArrayList<Put>();
for (String labelExp : labelExps) {
Put put = new Put(Bytes.toBytes("row" + i));
put.add(fam, qual, HConstants.LATEST_TIMESTAMP, value);
puts.add(put);
i++;
}
table.put(puts);
} finally {
if (table != null) {
table.flushCommits();
}
}
...

Using the Apache Thrift Proxy API


The Apache Thrift library provides cross-language client-server remote procedure calls (RPCs), using Thrift
bindings. A Thrift binding is client code generated by the Apache Thrift Compiler for a target language (such as
Python) that allows communication between the Thrift server and clients using that client code. HBase includes
an Apache Thrift Proxy API, which allows you to write HBase applications in Python, C, C++, or another language
that Thrift supports. The Thrift Proxy API is slower than the Java API and may have fewer features. T use the
Thrift Proxy API, you need to configure and run the HBase Thrift server on your cluster. See Installing and Starting
the HBase Thrift Server. You also need to install the Apache Thrift compiler on your development system.
After the Thrift server is configured and running, generate Thrift bindings for the language of your choice, using
an IDL file. A HBase IDL file named HBase.thrift is included as part of HBase. After generating the bindings,
copy the Thrift libraries for your language into the same directory as the generated bindings. In the following
Python example, these libraries provide the thrift.transport and thrift.protocol libraries. These commands
show how you might generate the Thrift bindings for Python and copy the libraries on a Linux system.
$
$
$
$
$
$
$

mkdir HBaseThrift
cd HBaseThrift/
thrift -gen py /path/to/Hbase.thrift
mv gen-py/* .
rm -rf gen-py/
mkdir thrift
cp -rp ~/Downloads/thrift-0.9.0/lib/py/src/* ./thrift/

The following iexample shows a simple Python application using the Thrift Proxy API.
from
from
from
from

thrift.transport import TSocket


thrift.protocol import TBinaryProtocol
thrift.transport import TTransport
hbase import Hbase

# Connect to HBase Thrift server


transport = TTransport.TBufferedTransport(TSocket.TSocket(host, port))
protocol = TBinaryProtocol.TBinaryProtocolAccelerated(transport)
# Create and open the client connection
client = Hbase.Client(protocol)
transport.open()
# Modify a single row

114 | Cloudera Administration

Managing CDH and Managed Services


mutations = [Hbase.Mutation(
column='columnfamily:columndescriptor', value='columnvalue')]
client.mutateRow('tablename', 'rowkey', mutations)
# Modify a batch of rows
# Create a list of mutations per work of Shakespeare
mutationsbatch = []
for line in myDataFile:
rowkey = username + "-" + filename + "-" + str(linenumber).zfill(6)
mutations = [
Hbase.Mutation(column=messagecolumncf, value=line.strip()),
Hbase.Mutation(column=linenumbercolumncf, value=encode(linenumber)),
Hbase.Mutation(column=usernamecolumncf, value=username)
]
mutationsbatch.append(Hbase.BatchMutation(row=rowkey,mutations=mutations))
# Run the mutations for all the lines in myDataFile
client.mutateRows(tablename, mutationsbatch)
transport.close()

The Thrift Proxy API does not support writing to HBase clusters that are secured using Kerberos.
This example was modified from the following two blog posts on http://www.cloudera.com. See them for more
details.
Using the HBase Thrift Interface, Part 1
Using the HBase Thrift Interface, Part 2
Using the REST Proxy API
After configuring and starting the HBase REST Server on your cluster, you can use the HBase REST Proxy API to
stream data into HBase, from within another application or shell script, or by using an HTTP client such as wget
or curl. The REST Proxy API is slower than the Java API and may have fewer features. This approach is simple
and does not require advanced development experience to implement. However, like the Java and Thrift Proxy
APIs, it uses the full write path and can cause compactions and region splits.
Specified addresses without existing data create new values. Specified addresses with existing data create new
versions, overwriting an existing version if the row, column:qualifier, and timestamp all match that of the existing
value.
$ curl -H "Content-Type: text/xml" http://localhost:8000/test/testrow/test:testcolumn

The REST Proxy API does not support writing to HBase clusters that are secured using Kerberos.
For full documentation and more examples, see the REST Proxy API documentation.
Using Flume
Apache Flume is a fault-tolerant system designed for ingesting data into HDFS, for use with Hadoop. You can
configure Flume to write data directly into HBase. Flume includes two different sinks designed to work with
HBase: HBaseSink (org.apache.flume.sink.hbase.HBaseSink) and AsyncHBaseSink
(org.apache.flume.sink.hbase.AsyncHBaseSink). HBaseSink supports HBase IPC calls introduced in HBase 0.96,
and allows you to write data to an HBase cluster that is secured by Kerberos, whereas AsyncHBaseSink does
not. However, AsyncHBaseSink uses an asynchronous model and guarantees atomicity at the row level.
You configure HBaseSink and AsyncHBaseSink nearly identically. Following is an example configuration for each.
Bold lines highlight differences in the configurations. For full documentation about configuring HBaseSink and
AsyncHBaseSink, see the Flume documentation. The table, columnFamily, and column parameters correlate
to the HBase table, column family, and column where the data is to be imported. The serializer is the class that
converts the data at the source into something HBase can use. Configure your sinks in the Flume configuration
file.

Cloudera Administration | 115

Managing CDH and Managed Services


In practice, you usually need to write your own serializer, which implements either AsyncHBaseEventSerializer
or HBaseEventSerializer. The HBaseEventSerializer converts Flume Events into one or more HBase Puts, sends
them to the HBase cluster, and is closed when the HBaseSink stops. AsyncHBaseEventSerializer starts and
listens for Events. When it receives an Event, it calls the setEvent method and then calls the getActions and
getIncrements methods. When the AsyncHBaseSink is stopped, the serializer cleanUp method is called. These
methods return PutRequest and AtomicIncrementRequest, which are part of the asynchbase API.
AsyncHBaseSink:
#Use the AsyncHBaseSink
host1.sinks.sink1.type = org.apache.flume.sink.hbase.AsyncHBaseSink
host1.sinks.sink1.channel = ch1
host1.sinks.sink1.table = transactions
host1.sinks.sink1.columnFamily = clients
host1.sinks.sink1.column = charges
host1.sinks.sink1.batchSize = 5000
#Use the SimpleAsyncHbaseEventSerializer that comes with Flume
host1.sinks.sink1.serializer =
org.apache.flume.sink.hbase.SimpleAsyncHbaseEventSerializer
host1.sinks.sink1.serializer.incrementColumn = icol
host1.channels.ch1.type=memory

HBaseSink:
#Use the HBaseSink
host1.sinks.sink1.type = org.apache.flume.sink.hbase.HBaseSink
host1.sinks.sink1.channel = ch1
host1.sinks.sink1.table = transactions
host1.sinks.sink1.columnFamily = clients
host1.sinks.sink1.column = charges
host1.sinks.sink1.batchSize = 5000
#Use the SimpleHbaseEventSerializer that comes with Flume
host1.sinks.sink1.serializer = org.apache.flume.sink.hbase.SimpleHbaseEventSerializer
host1.sinks.sink1.serializer.incrementColumn = icol
host1.channels.ch1.type=memory

The following serializer, taken from an Apache Flume blog post by Dan Sandler, splits the event body based on
a delimiter and inserts each split into a different column. The row is defined in the event header. When each
event is received, a counter is incremented to track the number of events received.
/**
* A serializer for the AsyncHBaseSink, which splits the event body into
* multiple columns and inserts them into a row whose key is available in
* the headers
*/
public class SplittingSerializer implements AsyncHbaseEventSerializer {
private byte[] table;
private byte[] colFam;
private Event currentEvent;
private byte[][] columnNames;
private final List<PutRequest> puts = new ArrayList<PutRequest>();
private final List<AtomicIncrementRequest> incs = new
ArrayList<AtomicIncrementRequest>();
private byte[] currentRowKey;
private final byte[] eventCountCol = "eventCount".getBytes();
@Override
public void initialize(byte[] table, byte[] cf) {
this.table = table;
this.colFam = cf;
}
@Override
public void setEvent(Event event) {
// Set the event and verify that the rowKey is not present
this.currentEvent = event;
String rowKeyStr = currentEvent.getHeaders().get("rowKey");
if (rowKeyStr == null) {
throw new FlumeException("No row key found in headers!");
}
currentRowKey = rowKeyStr.getBytes();
}
@Override
public List<PutRequest> getActions() {

116 | Cloudera Administration

Managing CDH and Managed Services


// Split the event body and get the values for the columns
String eventStr = new String(currentEvent.getBody());
String[] cols = eventStr.split(",");
puts.clear();
for (int i = 0; i < cols.length; i++) {
//Generate a PutRequest for each column.
PutRequest req = new PutRequest(table, currentRowKey, colFam,
columnNames[i], cols[i].getBytes());
puts.add(req);
}
return puts;
}
@Override
public List<AtomicIncrementRequest> getIncrements() {
incs.clear();
//Increment the number of events received
incs.add(new AtomicIncrementRequest(table, "totalEvents".getBytes(), colFam,
eventCountCol));
return incs;
}
@Override
public void cleanUp() {
table = null;
colFam = null;
currentEvent = null;
columnNames = null;
currentRowKey = null;
}
@Override
public void configure(Context context) {
//Get the column names from the configuration
String cols = new String(context.getString("columns"));
String[] names = cols.split(",");
byte[][] columnNames = new byte[names.length][];
int i = 0;
for(String name : names) {
columnNames[i++] = name.getBytes();
}
}
@Override
public void configure(ComponentConfiguration conf) {
}
}

Using Spark
You can write data to HBase from Apache Spark by using def saveAsHadoopDataset(conf: JobConf):
Unit. This example is adapted from a post on the spark-users mailing list.
// Note: mapred package is used, instead of the
// mapreduce package which contains new hadoop APIs.
import org.apache.hadoop.hbase.mapred.TableOutputFormat
import org.apache.hadoop.hbase.client
// ... some other settings
val conf = HBaseConfiguration.create()
// general hbase settings
conf.set("hbase.rootdir",
"hdfs://" + nameNodeURL + ":" + hdfsPort + "/hbase")
conf.setBoolean("hbase.cluster.distributed", true)
conf.set("hbase.zookeeper.quorum", hostname)
conf.setInt("hbase.client.scanner.caching", 10000)
// ... some other settings
val jobConfig: JobConf = new JobConf(conf, this.getClass)
// Note: TableOutputFormat is used as deprecated code
// because JobConf is an old hadoop API
jobConfig.setOutputFormat(classOf[TableOutputFormat])
jobConfig.set(TableOutputFormat.OUTPUT_TABLE, outputTable)

Cloudera Administration | 117

Managing CDH and Managed Services


Next, provide the mapping between how the data looks in Spark and how it should look in HBase. The following
example assumes that your HBase table has two column families, col_1 and col_2, and that your data is formatted
in sets of three in Spark, like (row_key, col_1, col_2).
def convert(triple: (Int, Int, Int)) = {
val p = new Put(Bytes.toBytes(triple._1))
p.add(Bytes.toBytes("cf"),
Bytes.toBytes("col_1"),
Bytes.toBytes(triple._2))
p.add(Bytes.toBytes("cf"),
Bytes.toBytes("col_2"),
Bytes.toBytes(triple._3))
(new ImmutableBytesWritable, p)
}

To write the data from Spark to HBase, you might use:


new PairRDDFunctions(localData.map(convert)).saveAsHadoopDataset(jobConfig)

Using Spark and Kafka


This example, written in Scala, uses Apache Spark in conjunction with the Apache Kafka message bus to stream
data from Spark to HBase. The example was provided in SPARK-944. It produces some random words and then
stores them in an HBase table, creating the table if necessary.
package org.apache.spark.streaming.examples
import java.util.Properties
import kafka.producer._
import
}
import
import
import
import
import
import
import
import
import
import
import

org.apache.hadoop.hbase.{ HBaseConfiguration, HColumnDescriptor, HTableDescriptor


org.apache.hadoop.hbase.client.{ HBaseAdmin, Put }
org.apache.hadoop.hbase.io.ImmutableBytesWritable
org.apache.hadoop.hbase.mapred.TableOutputFormat
org.apache.hadoop.hbase.mapreduce.TableInputFormat
org.apache.hadoop.hbase.util.Bytes
org.apache.hadoop.mapred.JobConf
org.apache.spark.SparkContext
org.apache.spark.rdd.{ PairRDDFunctions, RDD }
org.apache.spark.streaming._
org.apache.spark.streaming.StreamingContext._
org.apache.spark.streaming.kafka._

object MetricAggregatorHBase {
def main(args : Array[String]) {
if (args.length < 6) {
System.err.println("Usage: MetricAggregatorTest <master> <zkQuorum> <group>
<topics> <destHBaseTableName> <numThreads>")
System.exit(1)
}
val Array(master, zkQuorum, group, topics, hbaseTableName, numThreads) = args
val conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", zkQuorum)
// Initialize hBase table if necessary
val admin = new HBaseAdmin(conf)
if (!admin.isTableAvailable(hbaseTableName)) {
val tableDesc = new HTableDescriptor(hbaseTableName)
tableDesc.addFamily(new HColumnDescriptor("metric"))
admin.createTable(tableDesc)
}
// setup streaming context
val ssc = new StreamingContext(master, "MetricAggregatorTest", Seconds(2),
System.getenv("SPARK_HOME"), StreamingContext.jarOfClass(this.getClass))
ssc.checkpoint("checkpoint")

118 | Cloudera Administration

Managing CDH and Managed Services


val topicpMap = topics.split(",").map((_, numThreads.toInt)).toMap
val lines = KafkaUtils.createStream(ssc, zkQuorum, group, topicpMap)
.map { case (key, value) => ((key, Math.floor(System.currentTimeMillis() /
60000).toLong * 60), value.toInt) }
val aggr = lines.reduceByKeyAndWindow(add _, Minutes(1), Minutes(1), 2)
aggr.foreach(line => saveToHBase(line, zkQuorum, hbaseTableName))
ssc.start
ssc.awaitTermination
}
def add(a : Int, b : Int) = { (a + b) }
def saveToHBase(rdd : RDD[((String, Long), Int)], zkQuorum : String, tableName :
String) = {
val conf = HBaseConfiguration.create()
conf.set("hbase.zookeeper.quorum", zkQuorum)
val jobConfig = new JobConf(conf)
jobConfig.set(TableOutputFormat.OUTPUT_TABLE, tableName)
jobConfig.setOutputFormat(classOf[TableOutputFormat])
new PairRDDFunctions(rdd.map { case ((metricId, timestamp), value) =>
createHBaseRow(metricId, timestamp, value) }).saveAsHadoopDataset(jobConfig)
}
def createHBaseRow(metricId : String, timestamp : Long, value : Int) = {
val record = new Put(Bytes.toBytes(metricId + "~" + timestamp))
record.add(Bytes.toBytes("metric"), Bytes.toBytes("col"),
Bytes.toBytes(value.toString))
(new ImmutableBytesWritable, record)
}
}
// Produces some random words between 1 and 100.
object MetricDataProducer {
def main(args : Array[String]) {
if (args.length < 2) {
System.err.println("Usage: MetricDataProducer <metadataBrokerList> <topic>
<messagesPerSec>")
System.exit(1)
}
val Array(brokers, topic, messagesPerSec) = args
// ZooKeeper connection properties
val props = new Properties()
props.put("metadata.broker.list", brokers)
props.put("serializer.class", "kafka.serializer.StringEncoder")
val config = new ProducerConfig(props)
val producer = new Producer[String, String](config)
// Send some messages
while (true) {
val messages = (1 to messagesPerSec.toInt).map { messageNum =>
{
val metricId = scala.util.Random.nextInt(10)
val value = scala.util.Random.nextInt(1000)
new KeyedMessage[String, String](topic, metricId.toString, value.toString)
}
}.toArray
producer.send(messages : _*)
Thread.sleep(100)

Cloudera Administration | 119

Managing CDH and Managed Services


}
}
}

Using a Custom MapReduce Job


Many of the methods to import data into HBase use MapReduce implicitly. If none of those approaches fit your
needs, you can use MapReduce directly to convert data to a series of HFiles or API calls for import into HBase.
In this way, you can import data from Avro, Parquet, or another format into HBase, or export data from HBase
into another format, using API calls such as TableOutputFormat, HFileOutputFormat, and TableInputFormat.

Configuring HBase MultiWAL Support


CDH supports multiple write-ahead logs (MultiWAL) for HBase. (For more information, see HBASE-5699.)
Without MultiWAL support, each region on a RegionServer writes to the same WAL. A busy RegionServer might
host several regions, and each write to the WAL is serial because HDFS only supports sequentially written files.
This causes the WAL to negatively impact performance.
MultiWAL allows a RegionServer to write multiple WAL streams in parallel by using multiple pipelines in the
underlying HDFS instance, which increases total throughput during writes.
Note: In the current implementation of MultiWAL, incoming edits are partitioned by Region. Therefore,
throughput to a single Region is not increased.
To configure MultiWAL for a RegionServer, set the value of the property hbase.wal.provider to multiwal and
restart the RegionServer. To disable MultiWAL for a RegionServer, unset the property and restart the RegionServer.
RegionServers using the original WAL implementation and those using the MultiWAL implementation can each
handle recovery of either set of WALs, so a zero-downtime configuration update is possible through a rolling
restart.
Configuring MultiWAL Support Using Cloudera Manager
1.
2.
3.
4.
5.
6.
7.
8.

Go to the HBase service.


Click the Configuration tab.
Select Scope > RegionServer.
Select Category > Main.
Set WAL Provider to MultiWAL.
Set the Per-RegionServer Number of WAL Pipelines to a value greater than 1.
Click Save Changes to commit the changes.
Restart the RegionServer roles.

Configuring MultiWAL Support Using the Command Line


Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Edit hbase-site.xml on each RegionServer where you want to enable MultiWAL. Add the following property
by pasting the XML.
<property>
<name>hbase.wal.provider</name>
<value>multiwal</value>
</property>

2. Stop and restart the RegionServer.


120 | Cloudera Administration

Managing CDH and Managed Services


Storing Medium Objects (MOBs) in HBase
Data comes in many sizes, and saving all of your data in HBase, including binary data such as images and
documents, is convenient. HBase can technically handle binary objects with cells that are up to 10MB in size.
However, HBase's normal read and write paths are optimized for values smaller than 100KB in size. When HBase
deals with large numbers of values up to 10MB, referred to here as medium objects, or MOBs, performance is
degraded due to write amplification caused by splits and compactions.
Traditionally, the way to solve this problem has been to store objects larger than 100KB directly in HDFS, and
store references to their locations in HBase. CDH 5.4 introduces optimizations for storing objects up to 10MB
in size, termed medium objects or MOBs, directly in HBase, via the work done in HBASE-11339.
To take advantage of MOB, you need to use HFile version 3. Optionally, you can configure the MOB file reader's
cache settings for each RegionServer, then configure specific columns to hold MOB data. Client code does not
need to change to take advantage of HBase MOB support. The feature is transparent to the client.
Enabling HFile Version 3 Using Cloudera Manager
Required Role:
To enable HFile version 3 using Cloudera Manager, edit the HBase RegionServer advanced configuration snippet.
1.
2.
3.
4.

Go to the HBase service.


Click the Configuration tab.
Search for the property HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml.
Paste the following XML into the Value field and save your changes.
<property>
<name>hfile.format.version</name>
<value>3</value>
</property>

Changes will take effect after the next major compaction.


Enabling HFile Version 3 Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Paste the following XML into hbase-site.xml.
<property>
<name>hfile.format.version</name>
<value>3</value>
</property>

Restart HBase. Changes will take effect for a given region during its next major compaction.
Configuring Columns to Store MOBs
Two configuration options are provided to configure a column to store MOBs:
IS_MOB is a Boolean option, which specifies whether or not the column can store MOBs.
MOB_THRESHOLD configures the number of bytes at which an object is considered to be a MOB. If you do not
specify a value for MOB_THRESHOLD, the default is 100 KB. If you write a value larger than this threshold, it is
treated as a MOB.
You can configure a column to store MOBs using the HBase Shell or the Java API.
Cloudera Administration | 121

Managing CDH and Managed Services


Using HBase Shell:
hbase> create 't1', {NAME => 'f1', IS_MOB => true, MOB_THRESHOLD => 102400}
hbase> alter 't1', {NAME => 'f1', IS_MOB => true, MOB_THRESHOLD =>
102400}

Using the Java API:


HColumnDescriptor hcd = new HColumnDescriptor(f);
hcd.setMobEnabled(true);
hcd.setMobThreshold(102400L);

HBase MOB Cache Properties


Because there can be a large number of MOB files at any time, as compared to the number of HFiles, MOB files
are not always kept open. The MOB file reader cache is a LRU cache which keeps the most recently used MOB
files open.
The following properties are available for tuning the HBase MOB cache.
Table 2: HBase MOB Cache Properties
Property

Default

Description

hbase.mob.file.cache.size

1000

The of opened file handlers to cache.


A larger value will benefit reads by
providing more file handlers per
MOB file cache and would reduce
frequent file opening and closing of
files. However, if the value is too
high, errors such as "Too many
opened file handlers" may be logged.

hbase.mob.cache.evict.period 3600

The amount of time in seconds after


a file is opened before the MOB
cache evicts cached files. The default
value is 3600 seconds.

hbase.mob.cache.evict.remain.ratio 0.5f

The ratio, expressed as a float


between 0.0 and 1.0, that controls
how manyfiles remain cached after
an eviction is triggered due to the
number of cached files exceeding
the hbase.mob.file.cache.size.
The default value is 0.5f.

Configuring the MOB Cache Using Cloudera Manager


To configure the MOB cache within Cloudera Manager, edit the HBase Service advanced aonfiguration Snippet
for the cluster. Cloudera recommends testing your configuration with the default settings first.
1.
2.
3.
4.

Go to the HBase service.


Click the Configuration tab.
Search for the property HBase Service Advanced Configuration Snippet (Safety Valve) for hbase-site.xml.
Paste your configuration into the Value field and save your changes. The following example sets the
hbase.mob.cache.evict.period property to 5000 seconds. See Table 2: HBase MOB Cache Properties on
page 122 for a full list of configurable properties for HBase MOB.
<property>
<name>hbase.mob.cache.evict.period</name>

122 | Cloudera Administration

Managing CDH and Managed Services


<value>5000</value>
</property>

5. Restart your cluster for the changes to take effect.


Configuring the MOB Cache Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Because there can be a large number of MOB files at any time, as compared to the number of HFiles, MOB files
are not always kept open. The MOB file reader cache is a LRU cache which keeps the most recently used MOB
files open.
To customize the configuration of the MOB file reader's cache on each RegionServer, configure the MOB cache
properties in the RegionServer's hbase-site.xml. Customize the configuration to suit your environment, and
restart or rolling restart the RegionServer. Cloudera recommends testing your configuration with the default
settings first. The following example sets the hbase.mob.cache.evict.period property to 5000 seconds. See
Table 2: HBase MOB Cache Properties on page 122 for a full list of configurable properties for HBase MOB.
<property>
<name>hbase.mob.cache.evict.period</name>
<value>5000</value>
</property>

Testing MOB Storage and Retrieval Performance


HBase provides the Java utility org.apache.hadoop.hbase.IntegrationTestIngestMOB to assist with testing
the MOB feature and deciding on appropriate configuration values for your situation. The utility is run as follows:
$ sudo -u hbase hbase org.apache.hadoop.hbase.IntegrationTestIngestMOB \
-threshold 102400 \
-minMobDataSize 512 \
-maxMobDataSize 5120

threshold is the threshold at which cells are considered to be MOBs. The default is 1 kB, expressed in bytes.
minMobDataSize is the minimum value for the size of MOB data. The default is 512 B, expressed in bytes.
maxMobDataSize is the maximum value for the size of MOB data. The default is 5 kB, expressed in bytes.
Compact MOB Files Manually
You can trigger manual compaction of MOB files manually, rather than waiting for them to be triggered by your
configuration, using the HBase Shell commands compact_mob and major_compact_mob. Each of these commands
requires the first parameter to be the table name, and takes an optional column family name as the second
argument. If the column family is provided, only that column family's files are compacted. Otherwise, all
MOB-enabled column families' files are compacted.
hbase>
hbase>
hbase>
hbase>

compact 't1'
compact 't1', 'f1'
major_compact 't1'
major_compact 't1', 'f1'

This functionality is also available via the API, using the Admin.compact and Admin.majorCompact methods.

Exposing HBase Metrics to a Ganglia Server


Ganglia is a popular open-source monitoring framework. You can expose HBase metrics to a Ganglia instance
so that Ganglia can detect potential problems with your HBase cluster.

Cloudera Administration | 123

Managing CDH and Managed Services


Expose HBase Metrics to Ganglia Using Cloudera Manager
Required Role:
1. Go to the HBase service.
2. Click the Configuration tab.
3. Select the HBase Master or RegionServer role. To monitor both, configure each role as described in the rest
of the procedure.
4. Select Category > Metrics.
5. Locate the Hadoop Metrics2 Advanced Configuration Snippet (Safety Valve) property or search for it by typing
its name in the Search box.
6. Edit the property. Add the following, substituting the server information with your own.
hbase.sink.ganglia.class=org.apache.hadoop.metrics2.sink.ganglia.GangliaSink31
hbase.sink.ganglia.servers=<Ganglia server>:<port>
hbase.sink.ganglia.period=10

If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Restart the role.
9. Restart the service.
Expose HBase Metrics to Ganglia Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Edit /etc/hbase/conf/hadoop-metrics2-hbase.properties on the master or RegionServers you want
to monitor, and add the following properties, substituting the server information with your own:
hbase.sink.ganglia.class=org.apache.hadoop.metrics2.sink.ganglia.GangliaSink31
hbase.sink.ganglia.servers=<Ganglia server>:<port>
hbase.sink.ganglia.period=10

2. Restart the master or RegionServer.

Managing HDFS
The section contains configuration tasks for the HDFS service. For information on configuring HDFS for high
availability, see HDFS High Availability on page 288.

Managing Federated Nameservices


Required Role:
Cloudera Manager supports the configuration of multiple nameservices managing separate HDFS namespaces,
all of which share the storage available on the set of DataNodes. These nameservices are federated, meaning
each nameservice is independent and does not require coordination with other nameservices. See HDFS Federation
for more information.
It is simplest to add a second nameservice if high availability is already enabled. The process of enabling high
availability creates a nameservice as part of the enable high availability workflow.

124 | Cloudera Administration

Managing CDH and Managed Services


Important: Configuring a new nameservice shut downs the services that depend upon HDFS. Once
the new nameservice has been started, the services that depend upon HDFS must be restarted, and
the client configurations must be redeployed. (This can be done as part of the Add Nameservice
workflow, as an option.)
Configuring the First Nameservice Using Cloudera Manager
Follow the instructions below to define the first nameservice.
1. Go to the HDFS service.
2. Click the Configuration tab.
3. Type nameservice in the Search field. The nameservice properties for the NameNode and SecondaryNameNode
display.
4. In the NameNode Nameservice field, type a name for the nameservice. The name must be unique and can
contain only alphanumeric characters.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
5. In the Mountpoints field, change the mount point from "/" to a list of mount points that are in the namespace
that this nameservice will manage. (You can enter this as a comma-separated list for example, /hbase,
/tmp, /user or by clicking the
icon to add each mount point in its own field.) You can determine the list
of mount points by running the command hadoop fs -ls / from the CLI on the NameNode host.
6. In the SecondaryNameNode Nameservice field, type the nameservice name that you provided for the
NameNode Nameservice property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Click the Instances tab. The Federation and High Availability section displays with the nameservice listed.
Editing the List of Mountpoints for a Nameservice Using Cloudera Manager
1. Go to the HDFS service.
2. Click the Instances tab. The Federation and High Availability section displays with the nameservices listed.
3. Select Actions > Edit. In the Mount Points field, change the mount point to a list of mount points in the
namespace that the nameservice will manage.
4. Click OK.
Adding a Nameservice Using Cloudera Manager
The instructions below for adding a nameservice assume that one nameservice is already set up. The first
nameservice cam be set up either by configuring the first nameservice or by enabling HDFS high availability.
1. Go to the HDFS service.
2. Click the Instances tab. At the top of this page you should see the Federation and High Availability section.
If this section does not appear, it means you do not have any nameservices configured. You must have one
nameservice already configured in order to add a nameservice.
3. Click the Add Nameservice button.
a. In the Nameservice Name field, enter a name for the new nameservice. The name must be unique and
can contain only alphanumeric characters.
b. In the Mount Points field, enter at least one mount point for the nameservice. This defines the portion of
HDFS that will be managed under the new nameservice. (Click the to the right of the field to add a new
mount point). You cannot use "/" as a mount point; you must specify HDFS directories by name.
The mount points must be unique for this nameservice; you cannot specify any of the same mount
points you have used for other nameservices.

Cloudera Administration | 125

Managing CDH and Managed Services


You can specify mount points that do not yet exist, and create the corresponding directories in a later
step in this procedure.
If you want to use a mount point previously associated with another nameservice you must first
remove that mount point from that service. You can do this using the Edit command from the Actions
menu for that nameservice, and later add the mount point to the new nameservice.
After you have brought up the new nameservice, you must create the directories that correspond with
the mount points you specified in the new namespace.
If a mount point corresponds to a directory that formerly was under a different nameservice, you must
also move any contents of that directory, if appropriate as described in step 8.
If an HBase service is set to depend on the federated HDFS service, edit the mount points of the
existing nameservice to reference:
HBase root directory (default /hbase)
MapReduce system directory (default /tmp/mapred/system)
MapReduce JobTracker staging root directory (default value /user).
c. If you want to configure high availability for the nameservice, leave the Highly Available checkbox checked.
d. Click Continue.
4. Select the hosts on which the new NameNode and Secondary NameNodes will be created. (These must be
hosts that are not already running other NameNode or SecondaryNameNode instances, and their /dfs/nn
and /dfs/snn directories should be empty if they exist. Click Continue.
5. Enter or confirm the directory property values (these will differ depending on whether you are enabling high
availability for this nameservice, or not).
6. Select the Start Dependent Services checkbox if you need to create directories or move data onto the new
nameservice. Leave this checked if you want the workflow to restart services and redeploy the client
configurations as the last steps in the workflow.
7. Click Continue. If the process finished successfully, click Finish. The new nameservice displays in the Federation
and High Availability section in the Instances tab of the HDFS service.
8. Create the directories you want under the new nameservice using the CLI:
a. To create a directory in the new namespace, use the command hadoop fs -mkdir
/nameservices/nameservice/directory where nameservice is the new nameservice you just created
and directory is the directory that corresponds to a mount point you specified.
b. To move data from one nameservice to another, use distcp or manual export/import. dfs -cp and dfs
-mv will not work.
c. Verify that the directories and data are where you expect them to be.
9. Restart the dependent services.
Note: The monitoring configurations at the HDFS level apply to all nameservices. If you have two
nameservices, it is not possible to disable a check on one but not the other. Likewise, it's not possible
to have different event thresholds for the two nameservices.
Also see Changing a Nameservice Name for Highly Available HDFS Using Cloudera Manager on page 314.
Nameservice and Quorum-based Storage
With Quorum-based Storage, JournalNodes are shared across nameservices. So, if JournalNodes are present in
an HDFS service, all nameservices will have Quorum-based Storage enabled. To override this:
The dfs.namenode.shared.edits.dir configuration of the two NameNodes of a high availability nameservice
should be configured to include the value of the dfs.namenode.name.dirs setting, or
The dfs.namenode.edits.dir configuration of the one NameNode of a non-high availability nameservice
should be configured to include the value of the dfs.namenode.name.dirs setting.

126 | Cloudera Administration

Managing CDH and Managed Services


NameNodes
NameNodes maintain the namespace tree for HDFS and a mapping of file blocks to DataNodes where the data
is stored. A simple HDFS cluster can have only one primary NameNode, supported by a secondary NameNode
that periodically compresses the NameNode edits log file that contains a list of HDFS metadata modifications.
This reduces the amount of disk space consumed by the log file on the NameNode, which also reduces the
restart time for the primary NameNode. A high availability cluster contains two NameNodes: active and standby.
Formatting the NameNode and Creating the /tmp Directory
Formatting the NameNode and Creating the /tmp Directory Using Cloudera Manager
Required Role:
When you add an HDFS service, the wizard automatically formats the NameNode and creates the /tmp directory
on HDFS. If you quit the wizard or it does not finish, you can format the NameNode and create the /tmp directory
outside the wizard by doing these steps:
1.
2.
3.
4.
5.
6.

Stop the HDFS service if it is running. See Starting, Stopping, and Restarting Services on page 39.
Click the Instances tab.
Click the NameNode role instance.
Select Actions > Format.
Start the HDFS service.
Select Actions > Create /tmp Directory.

Formatting the NameNode and Creating the /tmp Directory Using the Command Line
See Formatting the NameNode.
Backing Up and Restoring HDFS Metadata
Backing Up HDFS Metadata Using Cloudera Manager
HDFS metadata backups can be used to restore a NameNode when both NameNode roles have failed. In addition,
Cloudera recommends backing up HDFS metadata before a major upgrade.
Required Role:
This backup method requires you to shut down the cluster.
1. Note the active NameNode.
2. Stop the cluster. It is particularly important that the NameNode role process is not running so that you can
make a consistent backup.
3. Go to the HDFS service.
4. Click the Configuration tab.
5. In the Search field, search for "NameNode Data Directories" and note the value.
6. On the active NameNode host, back up the directory listed in the NameNode Data Directories property. If
more than one is listed, make a backup of one directory, since each directory is a complete copy. For example,
if the NameNode data directory is /data/dfs/nn, do the following as root:
# cd /data/dfs/nn
# tar -cvf /root/nn_backup_data.tar .

You should see output like this:


./
./current/
./current/fsimage
./current/fstime
./current/VERSION
./current/edits
./image/
./image/fsimage

Cloudera Administration | 127

Managing CDH and Managed Services


If there is a file with the extension lock in the NameNode data directory, the NameNode most likely is still
running. Repeat the steps, starting by shutting down the NameNode role.

Restoring HDFS Metadata From a Backup Using Cloudera Manager


The following process assumes a scenario where both NameNode hosts have failed and you must restore from
a backup.
1.
2.
3.
4.
5.
6.
7.

Remove the NameNode, JournalNode, and Failover Controller roles from the HDFS service.
Add the host on which the NameNode role will run.
Create the NameNode data directory, ensuring that the permissions, ownership, and group are set correctly.
Copy the backed up files to the NameNode data directory.
Add the NameNode role to the host.
Add the Secondary NameNode role to another host.
Enable high availability. If not all roles are started after the wizard completes, restart the HDFS service. Upon
startup, the NameNode reads the fsimage file and loads it into memory. If the JournalNodes are up and
running and there are edit files present, any edits newer than the fsimage are applied.

Moving NameNode Roles


This section describes two procedures for moving NameNode roles. Both procedures require cluster downtime.
If high availability is enabled for the NameNode, you can use a Cloudera Manager wizard to automate the
migration process. Otherwise you must manually delete and add the NameNode role to a new host.
After moving a NameNode, if you have a Hive and/or Impala service, perform the steps in NameNode
Post-Migration Steps on page 130.

Moving Highly Available NameNode, Failover Controller, and JournalNode Roles Using the Migrate Roles Wizard
Required Role:
The Migrate Roles wizard allows you to move roles of a highly available HDFS service from one host to another.
You can use it to move NameNode, JournalNode, and Failover Controller roles.
Note: Requirements and Limitations

This procedure requires cluster downtime.


The configuration of HDFS and services that depend on it must be valid.
The destination host must be commissioned and healthy.
The NameNode must be highly available using quorum-based storage.
Nameservice federation (multiple namespaces) is not supported.
HDFS automatic failover must be enabled, and the cluster must have a running ZooKeeper service.
On hosts running active and standby NameNodes, make sure you have backed up the data
directories.
On hosts running JournalNodes, make sure you have backed up the JournalNode edits directory.
If a Hue service is present in the cluster, its HDFS Web Interface Role property must refer to an
HttpFS role, not to a NameNode role.
A majority of configured JournalNode roles must be running.
The Failover Controller role that is not located on the source host must be running.
If the source host is not functioning properly, or is not reliably reachable, decommission the host
before running the wizard.

To run the Migrate Roles wizard:


1. If the host to which you want to move the NameNode is not in the cluster, follow the instructions in Adding
a Host to the Cluster on page 54 to add the host.
2. Go to the HDFS service.
3. Click the Instances tab.
128 | Cloudera Administration

Managing CDH and Managed Services


4. Click the Migrate Roles button.
5. Click the Source Host text field and specify the host running the roles to migrate. In the Search field optionally
enter hostnames to filter the list of hosts and click Search.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name

6.

7.
8.
9.

Select the checkboxes next to the desired host. The list of available roles to migrate displays. Deselect any
roles you don't want to migrate. When migrating a NameNode, the co-located Failover Controller must be
migrated as well.
Click the Destination Host text field and specify the host to which the roles will be migrated. On destination
hosts, indicate whether to delete data in the NameNode data directories and JournalNode edits directory. If
you choose not to delete data and such role data exists, the Migrate Roles command will not complete
successfully.
Acknowledge that the migration process incurs service unavailability by selecting the Yes, I am ready to
restart the cluster now checkbox.
Click Continue. The Command Progress screen displays listing each step in the migration process.
When the migration completes, click Finish.

Moving a NameNode to a Different Host Using Cloudera Manager


Required Role:
Note: This procedure requires cluster downtime.
1. If the host to which you want to move the NameNode is not in the cluster, follow the instructions in Adding
a Host to the Cluster on page 54 to add the host.
2. Stop all cluster services.
3. Make a backup of the dfs.name.dir directories on the existing NameNode host. Make sure you back up
the fsimage and edits files. They should be the same across all of the directories specified by the
dfs.name.dir property.
4. Copy the files you backed up from dfs.name.dir directories on the old NameNode host to the host where
you want to run the NameNode.
5. Go to the HDFS service.
6. Click the Instances tab.
7. Select the checkbox next to the NameNode role instance and then click the Delete button. Click Delete again
to confirm.
8. In the Review configuration changes page that appears, click Skip.
9. Click Add Role Instances to add a NameNode role instance.
10. Select the host where you want to run the NameNode and then click Continue.
11. Specify the location of the dfs.name.dir directories where you copied the data on the new host, and then
click Accept Changes.
12. Start cluster services. After the HDFS service has started, Cloudera Manager distributes the new configuration
files to the DataNodes, which will be configured with the IP address of the new NameNode host.

Cloudera Administration | 129

Managing CDH and Managed Services


NameNode Post-Migration Steps
After moving a NameNode, if you have a Hive and/or Impala service, perform the following steps:
1.
2.
3.
4.

Go to the Hive service.


Stop the Hive service.
Select Actions > Update Hive Metastore NameNodes.
If you have an Impala service, restart the Impala service or run an INVALIDATE METADATA query.

DataNodes
DataNodes store data in a Hadoop cluster and is the name of the daemon that manages the data. File data is
replicated on multiple DataNodes for reliability and so that localized computation can be executed near the data.
How NameNode Manages Blocks on a Failed DataNode
After a period without any heartbeats (which by default is 10.5 minutes), a DataNode is assumed to be failed.
The following describes how the NameNode manages block replication in such cases.
1. NameNode determines which blocks were on the failed DataNode.
2. NameNode locates other DataNodes with copies of these blocks.
3. The DataNodes with block copies are instructed to copy those blocks to other DataNodes to maintain the
configured replication factor.
4. Follow the procedure in Replacing a Disk on a DataNode Host on page 130 or Performing Disk Hot Swap for
DataNodes on page 132 to bring a repaired DataNode back online.
Replacing a Disk on a DataNode Host
Required Role:
For CDH 5.3 and higher, see Performing Disk Hot Swap for DataNodes on page 132.
If one of your DataNode hosts experiences a disk failure, follow this process to replace the disk:
1.
2.
3.
4.
5.

Stop managed services.


Decommission the DataNode role instance.
Replace the failed disk.
Recommission the DataNode role instance.
Run the HDFS fsck utility to validate the health of HDFS. The utility normally reports over-replicated blocks
immediately after a DataNode is reintroduced to the cluster, which is automatically corrected over time.
6. Start managed services.
Adding and Removing Storage Directories for DataNodes

Adding and Removing Storage Directories Using Cloudera Manager


Adding Storage Directories
Required Role:
1.
2.
3.
4.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > DataNode.
Add the new storage directory to the DataNode Data Directory property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

5. Click Save Changes to commit the changes.

130 | Cloudera Administration

Managing CDH and Managed Services


Removing Storage Directories
Required Role:
1.
2.
3.
4.
5.

Stop the cluster.


Go to the HDFS service.
Click the Configuration tab.
Select Scope > DataNode.
Remove the current directories and add new ones to the DataNode Data Directory property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

6. Click Save Changes to commit the changes.


7. Copy the contents under the old directory to the new directory.
8. Start the cluster.
Configuring Storage-Balancing for DataNodes
You can configure HDFS to distribute writes on each DataNode in a manner that balances out available storage
among that DataNode's disk volumes.
By default a DataNode writes new block replicas to disk volumes solely on a round-robin basis. You can configure
a volume-choosing policy that causes the DataNode to take into account how much space is available on each
volume when deciding where to place a new replica.
You can configure
how much DataNode volumes are allowed to differ in terms of bytes of free disk space before they are
considered imbalanced, and
what percentage of new block allocations will be sent to volumes with more available disk space than others.

Configuring Storage-Balancing for DataNodes Using Cloudera Manager


Required Role:
1.
2.
3.
4.
5.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > DataNode.
Select Category > Advanced.
Configure the following properties (you can use the Search box to locate the properties):
Property

Value

Description

dfs.datanode.
fsdataset.
volume.choosing.
policy

org.apache.hadoop.
Enables storage balancing among the DataNode's
hdfs.server.datanode.
volumes.
fsdataset.
AvailableSpaceVolumeChoosingPolicy

dfs.datanode.
available-spacevolume-choosingpolicy.balancedspace-threshold

10737418240 (default)

The amount by which volumes are allowed to


differ from each other in terms of bytes of free
disk space before they are considered
imbalanced. The default is 10737418240 (10 GB).
If the free space on each volume is within this
range of the other volumes, the volumes will be
considered balanced and block assignments will
be done on a pure round-robin basis.

dfs.datanode.
available-spacevolume-choosingpolicy.balanced-

0.75 (default)

What proportion of new block allocations will be


sent to volumes with more available disk space
than others. The allowable range is 0.0-1.0, but
Cloudera Administration | 131

Managing CDH and Managed Services


Property

Value

space-preferencefraction

Description
set it in the range 0.5 - 1.0 (that is, 50-100%), since
there should be no reason to prefer that volumes
with less available disk space receive more block
allocations.

If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the role.

Configuring Storage-Balancing for DataNodes Using the Command Line


See Configuring Storage-Balancing for DataNodes.
Performing Disk Hot Swap for DataNodes
This section describes how to replace HDFS disks without shutting down a DataNode. This is referred to as hot
swap.
Warning: Requirements and Limitations

Hot swap is supported for CDH 5.4 and higher.


Hot swap can only add disks with empty data directories.
Removing a disk does not move the data off the disk, which could potentially result in data loss.
Do not perform hot swap on multiple hosts at the same time.

Performing Disk Hot Swap for DataNodes Using Cloudera Manager


Required Role:
1. Configure data directories to remove the disk you are swapping out:
a.
b.
c.
d.
e.
f.
g.

Go to the HDFS service.


Click the Instances tab.
Click the affected DataNode.
Click the Configuration tab.
Select Scope > DataNode.
Select Category > Main.
Change the value of the DataNode Data Directory property to remove the directories that are mount
points for the disk you are removing.
Warning: Change the value of this property only for the specific DataNode instance where you
are planning to hot swap the disk. Do not edit the role group value for this property. Doing so
will cause data loss.

2.
3.
4.
5.

Click Save Changes to commit the changes.


Refresh the affected DataNode. Select Actions > Refresh Data Directories.
Remove the old disk and add the replacement disk.
Change the value of the DataNode Data Directory property to add back the directories that are mount points
for the disk you added.
6. Click Save Changes to commit the changes.
7. Refresh the affected DataNode. Select Actions > Refresh Data Directories.
8. Run the HDFS fsck utility to validate the health of HDFS.

132 | Cloudera Administration

Managing CDH and Managed Services


Performing Disk Hot Swap for DataNodes Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Use these instructions to perform hot swap of disks in a cluster that is not managed by Cloudera Manager
To add and remove disks:
1. If you are adding disks, format and mount them.
2. Change the value of dfs.datanode.data.dir in hdfs-site.xml on the DataNode to reflect the directories
that will be used from now on (add new points and remove obsolete ones). For more information, see the
instructions for DataNodes under Configuring Local Storage Directories.
3. Start the reconfiguration process:
If Kerberos is enabled:
$ kinit -kt /path/to/hdfs.keytab hdfs/<fully.qualified.domain.name@YOUR-REALM.COM>
&& dfsadmin -reconfig datanode HOST:PORT start

If Kerberos is not enabled:


$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT start

where HOST:PORT is the DataNode's dfs.datanode.ipc.address (or its hostname and the port specified
in dfs.datanode.ipc.address; for example dnhost1.example.com:5678)
To check on the progress of the reconfiguration, you can use the status option of the command; for example,
if Kerberos is not enabled:
$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT status

4. Once the reconfiguration is complete, unmount any disks you have removed from the configuration.
5. Run the HDFS fsck utility to validate the health of HDFS.
To perform maintenance on a disk:
1. Change the value of dfs.datanode.data.dir in hdfs-site.xml on the DataNode to exclude the mount
point directories that reside on the affected disk and reflect only the directories that will be used during the
maintenance window. For more information, see the instructions for DataNodes under Configuring Local
Storage Directories.
2. Start the reconfiguration process:
If Kerberos is enabled:
$ kinit -kt /path/to/hdfs.keytab hdfs/<fully.qualified.domain.name@YOUR-REALM.COM>
&& dfsadmin -reconfig datanode HOST:PORT start

If Kerberos is not enabled:


$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT start

where HOST:PORT is the DataNode's dfs.datanode.ipc.address, or its hostname and the port specified
in dfs.datanode.ipc.address.

Cloudera Administration | 133

Managing CDH and Managed Services


To check on the progress of the reconfiguration, you can use the status option of the command; for example,
if Kerberos is not enabled:
$ sudo -u hdfs hdfs dfsadmin -reconfig datanode HOST:PORT status

3.
4.
5.
6.
7.
8.

Once the reconfiguration is complete, unmount the disk.


Perform maintenance on the disk.
Remount the disk.
Change the value of dfs.datanode.data.dir again to reflect the original set of mount points.
Repeat step 2.
Run the HDFS fsck utility to validate the health of HDFS.

JournalNodes
High Availabilty clusters use JournalNodes to synchronize active and standy NameNodes. The active NameNode
writes to each JournalNode with changes, or "edits," to HDFS namespace metadata. During failover, the standby
NameNode applies all edits from the JournalNodes before promoting itself to the active state.
Moving the JournalNode Edits Directory
Moving the JournalNode Edits Directory for an Role Instance Using Cloudera Manager
To change the location of the edits directory for one JournalNode instance:
1. Reconfigure the JournalNode Edits Directory.
a.
b.
c.
d.
e.
f.

Go to the HDFS service in Cloudera Manager.


Click JournalNode under Status Summary.
Click the JournalNode link for the instance you are changing.
Click the Configuration tab.
Set dfs.journalnode.edits.dir to the path of the new jn directory.
Click Save Changes.

2. Move the location of the JournalNode (jn) directory at the command line:
a. Connect to host of the JournalNode.
b. Copy the JournalNode (jn) directory to its new location with the -a option to preserve permissions:
cp -a /<old_path_to_jn_dir>/jn /<new_path_to_jn_dir>/jn

c. Rename the old jn directory to avoid confusion:


mv /<old_path_to_jn_dir>/jn /<old_path_to_jn_dir>/jn_to_delete

3. Redeploy the HDFS client configuration:


a. Go to the HDFS service.
b. Select Actions > Deploy Client Configuration.
4. Perform a Rolling Restart on page 41 for HDFS by selecting Actions > Rolling Restart. Use the default settings.
5. From the command line, delete the old jn_to_delete directory.

Moving the JournalNode Edits Directory for a Role Group Using Cloudera Manager
To change the location of the edits directory for each JournalNode in the JournalNode Default Group:
1. Stop all services on the cluster in Cloudera Manager:
a. Go to the Cluster.
b. Select Actions > Stop.
2. Find the list of JournalNode hosts:
134 | Cloudera Administration

Managing CDH and Managed Services


a. Go to the HDFS service.
b. Click JournalNode under Status Summary.
3. Move the location of each JournalNode (jn) directory at the command line:
a. Connect to each host with a JournalNode.
b. Per host, copy the JournalNode (jn) directory to its new location with the -a option to preserve permissions:
cp -a /<old_path_to_jn_dir>/jn /<new_path_to_jn_dir>/jn

c. Per host, rename the old jn directory to avoid confusion:


mv /<old_path_to_jn_dir>/jn /<old_path_to_jn_dir>/jn_to_delete

4. Reconfigure the JournalNode Default Group:


a.
b.
c.
d.
e.

Go to the HDFS service.


Click the Configuration tab.
Click JournalNode under Scope.
Set dfs.journalnode.edits.dir to the path of the new jn directory for all JournalNodes in the group.
Click Save Changes.

5. Redeploy the client configuration for the cluster:


a. Go to the Cluster.
b. Select Actions > Deploy Client Configuration.
6. Start all services on the cluster by selecting Actions > Start.
7. Delete the old jn_to_delete directories from the command line.
Moving JournalNodes Across Hosts
To move JournalNodes to a new host, see Moving Highly Available NameNode, Failover Controller, and JournalNode
Roles Using the Migrate Roles Wizard on page 128.

Configuring Short-Circuit Reads


So-called "short-circuit" reads bypass the DataNode, allowing a client to read the file directly, as long as the
client is co-located with the data. Short-circuit reads provide a substantial performance boost to many applications
and help improve HBase random read profile and Impala performance.
Short-circuit reads require libhadoop.so (the Hadoop Native Library) to be accessible to both the server and
the client. libhadoop.so is not available if you have installed from a tarball. You must install from an .rpm,
.deb, or parcel in order to use short-circuit local reads.
Configuring Short-Circuit Reads Using Cloudera Manager
Required Role:
Note: Short-circuit reads are enabled by default in Cloudera Manager.
1.
2.
3.
4.
5.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > Gateway or HDFS (Service-Wide).
Select Category > Performance.
Locate the Enable HDFS Short Circuit Read property or search for it by typing its name in the Search box.
Check the box to enable it.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
Cloudera Administration | 135

Managing CDH and Managed Services


6. Click Save Changes to commit the changes.
Configuring Short-Circuit Reads Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
Configure the following properties in hdfs-site.xml to enable short-circuit reads in a cluster that is not
managed by Cloudera Manager:
<property>
<name>dfs.client.read.shortcircuit</name>
<value>true</value>
</property>
<property>
<name>dfs.client.read.shortcircuit.streams.cache.size</name>
<value>1000</value>
</property>
<property>
<name>dfs.client.read.shortcircuit.streams.cache.expiry.ms</name>
<value>10000</value>
</property>
<property>
<name>dfs.domain.socket.path</name>
<value>/var/run/hadoop-hdfs/dn._PORT</value>
</property>

Note: The text _PORT appears just as shown; you do not need to substitute a number.
If /var/run/hadoop-hdfs/ is group-writable, make sure its group is root.

Configuring HDFS Trash


The Hadoop trash feature helps prevent accidental deletion of files and directories. If trash is enabled and a file
or directory is deleted using the Hadoop shell, the file is moved to the .Trash directory in the user's home
directory instead of being deleted. Deleted files are initially moved to the Current sub-directory of the .Trash
directory, and their original path is preserved. If trash checkpointing is enabled, the Current directory is periodically
renamed using a timestamp. Files in .Trash are permanently removed after a user-configurable time delay.
Files and directories in the trash can be restored simply by moving them to a location outside the .Trash
directory.
Important:
The trash feature is disabled by default. Cloudera recommends that you enable it on all production
clusters.
The trash feature works by default only for files and directories deleted using the Hadoop shell.
Files or directories deleted programmatically using other interfaces (WebHDFS or the Java APIs,
for example) are not moved to trash, even if trash is enabled, unless the program has implemented
a call to the trash functionality. (Hue, for example, implements trash as of CDH 4.4.)
Users can bypass trash when deleting files using the shell by specifying the -skipTrash option
to the hadoop fs -rm -r command. This can be useful when it is necessary to delete files that
are too large for the user's quota.

136 | Cloudera Administration

Managing CDH and Managed Services


Configuring HDFS Trash Using Cloudera Manager
Required Role:
Enabling and Disabling Trash
1.
2.
3.
4.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > Gateway.
Select or deselect the Use Trash checkbox.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

5. Click Save Changes to commit the changes.


6. Restart the cluster and deploy the cluster client configuration.
Setting the Trash Interval
1.
2.
3.
4.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > NameNode.
Specify the Filesystem Trash Interval property, which controls the number of minutes after which a trash
checkpoint directory is deleted and the number of minutes between trash checkpoints. For example, to enable
trash so that deleted files are deleted after 24 hours, set the value of the Filesystem Trash Interval property
to 1440.
Note: The trash interval is measured from the point at which the files are moved to trash, not
from the last time the files were modified.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

5. Click Save Changes to commit the changes.


6. Restart all NameNodes.

Configuring HDFS Trash Using the Command Line


See Enabling Trash.

HDFS Balancers
HDFS data might not always be distributed uniformly across DataNodes. One common reason is addition of
new DataNodes to an existing cluster. HDFS provides a balancer utility that analyzes block placement and
balances data across the DataNodes. The balancer moves blocks until the cluster is deemed to be balanced,
which means that the utilization of every DataNode (ratio of used space on the node to total capacity of the
node) differs from the utilization of the cluster (ratio of used space on the cluster to total capacity of the cluster)
by no more than a given threshold percentage. The balancer does not balance between individual volumes on
a single DataNode.
Configuring and Running the HDFS Balancer Using Cloudera Manager
Required Role:
In Cloudera Manager, the HDFS balancer utility is implemented by the Balancer role. The Balancer role usually
shows a health of None on the HDFS Instances tab because it does not run continuously.
The Balancer role is normally added (by default) when the HDFS service is installed. If it has not been added, you
must add a Balancer role in order to rebalance HDFS and to see the Rebalance action.

Cloudera Administration | 137

Managing CDH and Managed Services


Configuring the Balancer Threshold
The Balancer has a default threshold of 10%, which ensures that disk usage on each DataNode differs from the
overall usage in the cluster by no more than 10%. For example, if overall usage across all the DataNodes in the
cluster is 40% of the cluster's total disk-storage capacity, the script ensures that DataNode disk usage is between
30% and 50% of the DataNode disk-storage capacity. To change the threshold:
1.
2.
3.
4.
5.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > Balancer.
Select Category > Main.
Set the Rebalancing Threshold property.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

6. Click Save Changes to commit the changes.


Running the Balancer
1.
2.
3.
4.

Go to the HDFS service.


Ensure the service has a Balancer role.
Select Actions > Rebalance.
Click Rebalance to confirm. If you see a Finished status, the Balancer ran successfully.

Configuring and Running the HDFS Balancer Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
The HDFS balancer re-balances data across the DataNodes, moving blocks from over-utilized to under-utilized
nodes. As the system administrator, you can run the balancer from the command-line as necessary -- for
example, after adding new DataNodes to the cluster.
Points to note:
The balancer requires the capabilities of an HDFS superuser (for example, the hdfs user) to run.
The balancer does not balance between individual volumes on a single DataNode.
You can run the balancer without parameters, as follows:
sudo -u hdfs hdfs balancer

Note: If Kerberos is enabled, do not use commands in the form sudo -u <user> hadoop
<command>; they will fail with a security error. Instead, use the following commands: $ kinit
<user> (if you are using a password) or $ kinit -kt <keytab> <principal> (if you are using
a keytab) and then, for each command executed by this user, $ <command>
This runs the balancer with a default threshold of 10%, meaning that the script will ensure that disk usage
on each DataNode differs from the overall usage in the cluster by no more than 10%. For example, if overall
usage across all the DataNodes in the cluster is 40% of the cluster's total disk-storage capacity, the script
ensures that each DataNode's disk usage is between 30% and 50% of that DataNode's disk-storage capacity.
You can run the script with a different threshold; for example:
sudo -u hdfs hdfs balancer -threshold 5

138 | Cloudera Administration

Managing CDH and Managed Services


This specifies that each DataNode's disk usage must be (or will be adjusted to be) within 5% of the cluster's
overall usage.
You can adjust the network bandwidth used by the balancer, by running the dfsadmin
-setBalancerBandwidth command before you run the balancer; for example:
dfsadmin -setBalancerBandwidth newbandwidth

where newbandwidth is the maximum amount of network bandwidth, in bytes per second, that each DataNode
can use during the balancing operation. For more information about the bandwidth command, see
BalancerBandwidthCommand.
The balancer can take a long time to run, especially if you are running it for the first time, or do not run it
regularly.

Enabling WebHDFS
Enabling WebHDFS Using Cloudera Manager
Required Role:
To enable WebHDFS, proceed as follows:
1.
2.
3.
4.
5.
6.

Select the HDFS service.


Click the Configuration tab.
Select Scope > HDFS-1 (Service Wide)
Select the Enable WebHDFS property.
Click the Save Changes button.
Restart the HDFS service.

You can find a full explanation of the WebHDFS API in the WebHDFS API documentation.
Enabling WebHDFS Using the Command Line
See Enabling WebHDFS.

Adding HttpFS
Required Role:
Apache Hadoop HttpFS is a service that provides HTTP access to HDFS.
HttpFS has a REST HTTP API supporting all HDFS filesystem operations (both read and write).
Common HttpFS use cases are:
Read and write data in HDFS using HTTP utilities (such as curl or wget) and HTTP libraries from languages
other than Java (such as Perl).
Transfer data between HDFS clusters running different versions of Hadoop (overcoming RPC versioning
issues), for example using Hadoop DistCp.
Read and write data in HDFS in a cluster behind a firewall. (The HttpFS server acts as a gateway and is the
only system that is allowed to send and receive data through the firewall).
HttpFS supports Hadoop pseudo-authentication, HTTP SPNEGO Kerberos, and additional authentication
mechanisms via a plugin API. HttpFS also supports Hadoop proxy user functionality.
The webhdfs client file system implementation can access HttpFS via the Hadoop filesystem command (hadoop
fs), by using Hadoop DistCp, and from Java applications using the Hadoop file system Java API.
The HttpFS HTTP REST API is interoperable with the WebHDFS REST HTTP API.
For more information about HttpFS, see Hadoop HDFS over HTTP.
The HttpFS role is required for Hue when you enable HDFS high availability.

Cloudera Administration | 139

Managing CDH and Managed Services


Adding the HttpFS Role
1.
2.
3.
4.
5.
6.
7.

Go to the HDFS service.


Click the Instances tab.
Click Add Role Instances.
Click the text box below the HttpFS field. The Select Hosts dialog displays.
Select the host on which to run the role and click OK.
Click Continue.
Check the checkbox next to the HttpFS role and select Actions for Selected > Start.

Using Load Balancer with HttpFS


To configure a load balancer, select Clusters > HDFS > Configuration > Category > HttpFS and enter the hostname
and port number of the load balancer in the HTTPFS Load Balancer property in the format hostname:port
number.
Note:
When you set this property, Cloudera Manager regenerates the keytabs for HttpFS roles. The principal
in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this HDFS service, the Hue service has the option to use the
load balancer as its HDFS Web Interface Role.

Adding and Configuring an NFS Gateway


The NFSv3 gateway allows a client to mount HDFS as part of the client's local file system. The gateway machine
can be any host in the cluster, including the NameNode, a DataNode, or any HDFS client. The client can be any
NFSv3-client-compatible machine.
Important:
HDFS does not currently provide ACL support for an NFS gateway.
After mounting HDFS to his or her local filesystem, a user can:
Browse the HDFS file system as though it were part of the local file system
Upload and download files from the HDFS file system to and from the local file system.
Stream data directly to HDFS through the mount point.
File append is supported, but random write is not.
Adding and Configuring an NFS Gateway Using Cloudera Manager
Required Role:
The NFS Gateway role implements an NFSv3 gateway. It is an optional role for a CDH 5 HDFS service.
Requirements and Limitations
The NFS gateway works only with the following operating systems and Cloudera Manager and CDH versions:
With Cloudera Manager 5.0.1 or later and CDH 5.0.1 or later, the NFS gateway works on all operating
systems supported by Cloudera Manager.
With Cloudera Manager 5.0.0 or CDH 5.0.0, the NFS gateway only works on RHEL and similar systems.
The NFS gateway is not supported on versions earlier than Cloudera Manager 5.0.0 and CDH 5.0.0.
If any NFS server is already running on the NFS Gateway host, it must be stopped before the NFS Gateway
role is started.
140 | Cloudera Administration

Managing CDH and Managed Services


There are two configuration options related to NFS Gateway role: Temporary Dump Directory and Allowed
Hosts and Privileges. The Temporary Dump Directory is automatically created by the NFS Gateway role and
should be configured before starting the role.
The Access Time Precision property in the HDFS service must be enabled.
Adding and Configuring the NFS Gateway Role
1. Go to the HDFS service.
2. Click the Instances tab.
3. Click Add Role Instances.
4. Click the text box below the NFS Gateway field. The Select Hosts dialog displays.
5. Select the host on which to run the role and click OK.
6. Click Continue.
7. Click the NFS Gateway role.
8. Click the Configuration tab.
9. Select Scope > NFS Gateway.
10. Select Category > Main.
11. Ensure that the requirements on the directory set in the Temporary Dump Directory property are met.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
12. Optionally edit Allowed Hosts and Privileges.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
13. Click Save Changes to commit the changes.
14. Click the Instances tab.
15. Check the checkbox next to the NFS Gateway role and select Actions for Selected > Start.
Configuring an NFSv3 Gateway Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
The subsections that follow provide information on installing and configuring the gateway.
Note: Install Cloudera Repository
Before using the instructions on this page to install or upgrade, install the Cloudera yum, zypper/YaST
or apt repository, and install or upgrade CDH 5 and make sure it is functioning correctly. For
instructions, see Installing the Latest CDH 5 Release and Upgrading Unmanaged CDH Using the
Command Line.

Upgrading from a CDH 5 Beta Release


If you are upgrading from a CDH 5 Beta release, you must first remove the hadoop-hdfs-portmap package.
Proceed as follows.
1. Unmount existing HDFS gateway mounts. For example, on each client, assuming the file system is mounted
on /hdfs_nfs_mount:
$ umount /hdfs_nfs_mount

Cloudera Administration | 141

Managing CDH and Managed Services


2. Stop the services:
$ sudo service hadoop-hdfs-nfs3 stop
$ sudo hadoop-hdfs-portmap stop

3. Remove the hadoop-hdfs-portmap package.


On a RHEL-compatible system:
$ sudo yum remove hadoop-hdfs-portmap

On a SLES system:
$ sudo zypper remove hadoop-hdfs-portmap

On an Ubuntu or Debian system:


$ sudo apt-get remove hadoop-hdfs-portmap

4. Install the new version


On a RHEL-compatible system:
$ sudo yum install hadoop-hdfs-nfs3

On a SLES system:
$ sudo zypper install hadoop-hdfs-nfs3

On an Ubuntu or Debian system:


$ sudo apt-get install hadoop-hdfs-nfs3

5. Start the system default portmapper service:


$ sudo service portmap start

6. Now proceed with Starting the NFSv3 Gateway on page 144, and then remount the HDFS gateway mounts.

Installing the Packages for the First Time


On RHEL and similar systems:
Install the following packages on the cluster host you choose for NFSv3 Gateway machine (we'll refer to it as
the NFS server from here on).
nfs-utils
nfs-utils-lib
hadoop-hdfs-nfs3
The first two items are standard NFS utilities; the last is a CDH package.
Use the following command:
$ sudo yum install nfs-utils nfs-utils-lib hadoop-hdfs-nfs3

On SLES:

142 | Cloudera Administration

Managing CDH and Managed Services


Install nfs-utils on the cluster host you choose for NFSv3 Gateway machine (referred to as the NFS server
from here on):
$ sudo zypper install nfs-utils

On an Ubuntu or Debian system:


Install nfs-common on the cluster host you choose for NFSv3 Gateway machine (referred to as the NFS server
from here on):
$ sudo apt-get install nfs-common

Configuring the NFSv3 Gateway


Proceed as follows to configure the gateway.
1. Add the following property to hdfs-site.xml on the NameNode:
<property>
<name>dfs.namenode.accesstime.precision</name>
<value>3600000</value>
<description>The access time for an HDFS file is precise up to this value. The
default value is 1 hour.
Setting a value of 0 disables access times for HDFS.</description>
</property>

2. Add the following property to hdfs-site.xml on the NFS server:


<property>
<name>dfs.nfs3.dump.dir</name>
<value>/tmp/.hdfs-nfs</value>
</property>

Note:
You should change the location of the file dump directory, which temporarily saves out-of-order
writes before writing them to HDFS. This directory is needed because the NFS client often reorders
writes, and so sequential writes can arrive at the NFS gateway in random order and need to be
saved until they can be ordered correctly. After these out-of-order writes have exceeded 1MB in
memory for any given file, they are dumped to the dfs.nfs3.dump.dir (the memory threshold
is not currently configurable).
Make sure the directory you choose has enough space. For example, if an application uploads 10
files of 100MB each, dfs.nfs3.dump.dir should have roughly 1GB of free space to allow for a
worst-case reordering of writes to every file.
3. Configure the user running the gateway (normally the hdfs user as in this example) to be a proxy for other
users. To allow the hdfs user to be a proxy for all other users, add the following entries to core-site.xml
on the NameNode:
<property>
<name>hadoop.proxyuser.hdfs.groups</name>
<value>*</value>
<description>
Set this to '*' to allow the gateway user to proxy any group.
</description>
</property>
<property>
<name>hadoop.proxyuser.hdfs.hosts</name>
<value>*</value>
<description>
Set this to '*' to allow requests from any hosts to be proxied.

Cloudera Administration | 143

Managing CDH and Managed Services


</description>
</property>

4. Restart the NameNode.

Starting the NFSv3 Gateway


Do the following on the NFS server.
1. First, stop the default NFS services, if they are running:
$ sudo service nfs stop

2. Start the HDFS-specific services:


$ sudo service hadoop-hdfs-nfs3 start

Verifying that the NFSv3 Gateway is Working


To verify that the NFS services are running properly, you can use the rpcinfo command on any host on the
local network:
$ rpcinfo -p <nfs_server_ip_address>

You should see output such as the following:


program

vers

proto

port

100005
100005
100005
100000
100000
100005
100005
100003
100005

1
2
2
2
2
3
1
3
3

tcp
udp
tcp
tcp
udp
udp
udp
tcp
tcp

4242
4242
4242
111
111
4242
4242
2049
4242

mountd
mountd
mountd
portmapper
portmapper
mountd
mountd
nfs
mountd

To verify that the HDFS namespace is exported and can be mounted, use the showmount command.
$ showmount -e <nfs_server_ip_address>

You should see output similar to the following:


Exports list on <nfs_server_ip_address>:
/ (everyone)

Mounting HDFS on an NFS Client


To import the HDFS file system on an NFS client, use a mount command such as the following on the client:
$ mount -t

144 | Cloudera Administration

nfs

-o vers=3,proto=tcp,nolock <nfs_server_hostname>:/ /hdfs_nfs_mount

Managing CDH and Managed Services


Note:
When you create a file or directory as user hdfs on the client (that is, in the HDFS file system imported
via the NFS mount), the ownership may differ from what it would be if you had created it in HDFS
directly. For example, ownership of a file created on the client might be hdfs:hdfs when the same
operation done natively in HDFS resulted in hdfs:supergroup. This is because in native HDFS, BSD
semantics determine the group ownership of a newly-created file: it is set to the same group as the
parent directory where the file is created. When the operation is done over NFS, the typical Linux
semantics create the file with the group of the effective GID (group ID) of the process creating the file,
and this characteristic is explicitly passed to the NFS gateway and HDFS.

Setting HDFS Quotas


You can set quotas in HDFS for:
The number of file and directory names used
The amount of space used by given directories
Points to note:
The quotas for names and the quotas for space are independent of each other.
File and directory creation fails if the creation would cause the quota to be exceeded.
Allocation fails if the quota would prevent a full block from being written; keep this in mind if you are using
a large block size.
If you are using replication, remember that each replica of a block counts against the quota.
About file count limits
The file count quota is a limit on the number of file and directory names in the directory configured.
A directory counts against its own quota, so a quota of 1 forces the directory to remain empty.
File counts are based on the intended replication factor for the files; changing the replication factor for a file
will credit or debit quotas.
About disk space limits
The space quota is a hard limit on the number of bytes used by files in the tree rooted at the directory being
configured.
Each replica of a block counts against the quota.
The disk space quota calculation takes replication into account, so it uses the replicated size of each file, not
the user-facing size.
The disk space quota calculation includes open files (files presently being written), as well as files already
written.
Block allocations for files being written will fail if the quota would not allow a full block to be written.
Setting HDFS Quotas Using Cloudera Manager
Required Role:
1. From the HDFS service page, select the File Browser tab.
2. Browse the file system to find the directory for which you want to set quotas.
3. Click the directory name so that it appears in the gray panel above the listing of its contents and in the detail
section to the right of the File Browser table.
4. Click the Edit Quota button for the directory. A Manage Quota pop-up displays, where you can set file count
or disk space limits for the directory you have selected.
5. When you have set the limits you want, click OK.

Cloudera Administration | 145

Managing CDH and Managed Services


Setting HDFS Quotas Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
To set space quotas on a directory:
dfsadmin -setSpaceQuota n directory

where n is a number of bytes and directory is the directory the quota applies to. You can specify multiple
directories in a single command; napplies to each.
To remove space quotas from a directory:
dfsadmin -clrSpaceQuota directory

You can specify multiple directories in a single command.


To set name quotas on a directory:
dfsadmin -setQuota n directory

where n is the number of file and directory names in directory. You can specify multiple directories in a single
command; napplies to each.
To remove name quotas from a directory:
dfsadmin -clrQuota directory

You can specify multiple directories in a single command.


For More Information
For more information, see the HDFS Quotas Guide.

Configuring Mountable HDFS


CDH 5 includes a FUSE (Filesystem in Userspace) interface into HDFS. The hadoop-hdfs-fuse package enables
you to use your HDFS cluster as if it were a traditional filesystem on Linux. Proceed as follows.
Note: FUSE does not currently support file append operations.
Before you start: You must have a working HDFS cluster and know the hostname and port that your NameNode
exposes.
To install hadoop-hdfs-fuses On Red Hat-compatible systems:
$ sudo yum install hadoop-hdfs-fuse

To install hadoop-hdfs-fuse on Ubuntu systems:


$ sudo apt-get install hadoop-hdfs-fuse

To install hadoop-hdfs-fuse on SLES systems:


$ sudo zypper install hadoop-hdfs-fuse

You now have everything you need to begin mounting HDFS on Linux.
146 | Cloudera Administration

Managing CDH and Managed Services


To set up and test your mount point in a non-HA installation:
$ mkdir -p <mount_point>
$ hadoop-fuse-dfs dfs://<name_node_hostname>:<namenode_port> <mount_point>

where namenode_port is the NameNode's RPC port, dfs.namenode.servicerpc-address.


To set up and test your mount point in an HA installation:
$ mkdir -p <mount_point>
$ hadoop-fuse-dfs dfs://<nameservice_id> <mount_point>

where nameservice_id is the value of fs.defaultFS. In this case the port defined for
dfs.namenode.rpc-address.[nameservice ID].[name node ID] is used automatically. See Enabling HDFS
HA on page 291 for more information about these properties.
You can now run operations as if they are on your mount point. Press Ctrl+C to end the fuse-dfs program, and
umount the partition if it is still mounted.
Note:
To find its configuration directory, hadoop-fuse-dfs uses the HADOOP_CONF_DIR configured at the
time the mount command is invoked.
To clean up your test:
$ umount <mount_point>

You can now add a permanent HDFS mount which persists through reboots.
To add a system mount:
1. Open /etc/fstab and add lines to the bottom similar to these:
hadoop-fuse-dfs#dfs://<name_node_hostname>:<namenode_port> <mount_point> fuse
allow_other,usetrash,rw 2 0

For example:
hadoop-fuse-dfs#dfs://localhost:8020 /mnt/hdfs fuse allow_other,usetrash,rw 2 0

Note:
In an HA deployment, use the HDFS nameservice instead of the NameNode URI; that is, use the
value of dfs.nameservices in hdfs-site.xml.
2. Test to make sure everything is working properly:
$ mount <mount_point>

Your system is now configured to allow you to use the ls command and use that mount point as if it were a
normal system disk.
For more information, see the help for hadoop-fuse-dfs:
$ hadoop-fuse-dfs --help

Cloudera Administration | 147

Managing CDH and Managed Services


Optimizing Mountable HDFS
Cloudera recommends that you use the -obig_writes option on kernels later than 2.6.26. This option allows
for better performance of writes.
By default, the CDH 5 package installation creates the /etc/default/hadoop-fuse file with a maximum
heap size of 128 MB. You might need to change the JVM minimum and maximum heap size for better
performance. For example:
export LIBHDFS_OPTS="-Xms64m -Xmx256m"

Be careful not to set the minimum to a higher value than the maximum.

Configuring Centralized Cache Management in HDFS


Centralized cache management in HDFS is an explicit caching mechanism that allows users to specify paths to
be cached by HDFS. The NameNode will communicate with DataNodes that have the desired blocks on disk, and
instruct them to cache the blocks in off-heap caches.
This has several advantages:
Explicit pinning prevents frequently used data from being evicted from memory. This is particularly important
when the size of the working set exceeds the size of main memory, which is common for many HDFS
workloads.
Since DataNode caches are managed by the NameNode, applications can query the set of cached block
locations when making task placement decisions. Co-locating a task with a cached block replica improves
read performance.
When block has been cached by a DataNode, clients can use a new, more-efficient, zero-copy read API. Since
checksum verification of cached data is done once by the DataNode, clients can incur essentially zero overhead
when using this new API.
Centralized caching can improve overall cluster memory utilization. When relying on the OS buffer cache at
each DataNode, repeated reads of a block will result in all n replicas of the block being pulled into buffer
cache. With centralized cache management, you can explicitly pin only m of the n replicas, saving n-m memory.
Use Cases
Centralized cache management is most useful for files that are accessed repeatedly. For example, a fact table
in Hive which is often used in joins is a good candidate for caching. On the other hand, caching the input of a
one-year reporting query is probably less useful, since the historical data might be read only once.
Centralized cache management is also useful for mixed workloads with performance SLAs. Caching the working
set of a high-priority workload insures that it does not contend for disk I/O with a low-priority workload.

148 | Cloudera Administration

Managing CDH and Managed Services


Architecture

In this architecture, the NameNode is responsible for coordinating all the DataNode off-heap caches in the
cluster. The NameNode periodically receives a "cache report" from each DataNode which describes all the blocks
cached on a given DataNode. The NameNode manages DataNode caches by piggybacking cache and uncache
commands on the DataNode heartbeat.
The NameNode queries its set of cache directives to determine which paths should be cached. Cache directives
are persistently stored in the fsimage and edit log, and can be added, removed, and modified using Java and
command-line APIs. The NameNode also stores a set of cache pools, which are administrative entities used to
group cache directives together for resource management and enforcing permissions.
The NameNode periodically rescans the namespace and active cache directories to determine which blocks need
to be cached or uncached and assign caching to DataNodes. Rescans can also be triggered by user actions like
adding or removing a cache directive or removing a cache pool.
We do not currently cache blocks which are under construction, corrupt, or otherwise incomplete. If a cache
directive covers a symlink, the symlink target is not cached. Caching is currently done on a per-file basis, although
we would like to add block-level granularity in the future.
Concepts
Cache Directive
A cache directive defines a path that should be cached. Paths can be either directories or files. Directories are
cached non-recursively, meaning only files in the first-level listing of the directory will be cached. Directives also
specify additional parameters, such as the cache replication factor and expiration time. The replication factor
specifies the number of block replicas to cache. If multiple cache directives refer to the same file, the maximum
cache replication factor is applied.
The expiration time is specified on the command line as a time-to-live (TTL), a relative expiration time in the
future. After a cache directive expires, it is no longer considered by the NameNode when making caching decisions.

Cache Pool
A cache pool is an administrative entity used to manage groups of cache directives. Cache pools have UNIX-like
permissions that restrict which users and groups have access to the pool. Write permissions allow users to add
and remove cache directives to the pool. Read permissions allow users to list the cache directives in a pool, as
well as additional metadata. Execute permissions are unused.

Cloudera Administration | 149

Managing CDH and Managed Services


Cache pools are also used for resource management. Pools can enforce a maximum limit, which restricts the
number of bytes that can be cached in aggregate by directives in the pool. Normally, the sum of the pool limits
will approximately equal the amount of aggregate memory reserved for HDFS caching on the cluster. Cache
pools also track a number of statistics to help cluster users determine what is and should be cached.
Pools also enforce a maximum time-to-live. This restricts the maximum expiration time of directives being
added to the pool.
cacheadmin Command-Line Interface
On the command-line, administrators and users can interact with cache pools and directives via the hdfs
cacheadmin subcommand. Cache directives are identified by a unique, non-repeating 64-bit integer ID. IDs will
not be reused even if a cache directive is later removed. Cache pools are identified by a unique string name.

Cache Directive Commands


addDirective
Description: Add a new cache directive.
Usage: hdfs cacheadmin -addDirective -path <path> -pool <pool-name> [-force] [-replication
<replication>] [-ttl <time-to-live>]

Where, path: A path to cache. The path can be a directory or a file.


pool-name: The pool to which the directive will be added. You must have write permission on the cache pool in

order to add new directives.


force: Skips checking of cache pool resource limits.
replication: The cache replication factor to use. Defaults to 1.
time-to-live: Time period for which the directive is valid. Can be specified in seconds, minutes, hours, and
days, for example: 30m, 4h, 2d. The value never indicates a directive that never expires. If unspecified, the

directive never expires.


removeDirective
Description: Remove a cache directive.
Usage: hdfs cacheadmin -removeDirective <id>
Where, id: The id of the cache directive to remove. You must have write permission on the pool of the directive
in order to remove it. To see a list of PathBasedCache directive IDs, use the -listDirectives command.
removeDirectives
Description: Remove every cache directive with the specified path.
Usage: hdfs cacheadmin -removeDirectives <path>
Where, path: The path of the cache directives to remove. You must have write permission on the pool of the
directive in order to remove it.
listDirectives
Description: List PathBasedCache directives.
Usage: hdfs cacheadmin -listDirectives [-stats] [-path <path>] [-pool <pool>]
Where, path: List only PathBasedCache directives with this path. Note that if there is a PathBasedCache directive
for path in a cache pool that we don't have read access for, it will not be listed.
pool: List only path cache directives in that pool.
stats: List path-based cache directive statistics.

150 | Cloudera Administration

Managing CDH and Managed Services


Cache Pool Commands
addPool
Description: Add a new cache pool.
Usage: hdfs cacheadmin -addPool <name> [-owner <owner>] [-group <group>] [-mode <mode>]
[-limit <limit>] [-maxTtl <maxTtl>]

Where, name: Name of the new pool.


owner: Username of the owner of the pool. Defaults to the current user.
group: Group of the pool. Defaults to the primary group name of the current user.
mode: UNIX-style permissions for the pool. Permissions are specified in octal, for example: 0755. By default, this

is set to 0755.
limit: The maximum number of bytes that can be cached by directives in this pool, in aggregate. By default, no

limit is set.
maxTtl: The maximum allowed time-to-live for directives being added to the pool. This can be specified in

seconds, minutes, hours, and days, for example: 120s, 30m, 4h, 2d. By default, no maximum is set. A value of
never specifies that there is no limit.
modifyPool
Description: Modify the metadata of an existing cache pool.
Usage: hdfs cacheadmin -modifyPool <name> [-owner <owner>] [-group <group>] [-mode <mode>]
[-limit <limit>] [-maxTtl <maxTtl>]

Where, name: Name of the pool to modify.


owner: Username of the owner of the pool.
group: Groupname of the group of the pool.
mode: Unix-style permissions of the pool in octal.
limit: Maximum number of bytes that can be cached by this pool.
maxTtl: The maximum allowed time-to-live for directives being added to the pool.

removePool
Description: Remove a cache pool. This also uncaches paths associated with the pool.
Usage: hdfs cacheadmin -removePool <name>
Where, name: Name of the cache pool to remove.
listPools
Description: Display information about one or more cache pools, for example: name, owner, group, permissions,
and so on.
Usage: hdfs cacheadmin -listPools [-stats] [<name>]
Where, name: If specified, list only the named cache pool.
stats: Display additional cache pool statistics.

help
Description: Get detailed help about a command.
Cloudera Administration | 151

Managing CDH and Managed Services


Usage: hdfs cacheadmin -help <command-name>
Where, command-name: The command for which to get detailed help. If no command is specified, print detailed
help for all commands.
Configuration
Native Libraries
In order to lock block files into memory, the DataNode relies on native JNI code found in libhadoop.so. Be sure
to enable JNI if you are using HDFS centralized cache management.

Configuration Properties
Required
Be sure to configure the following in /etc/default/hadoop/conf/hdfs-default.xml:
dfs.datanode.max.locked.memory: The maximum amount of memory a DataNode will use for caching (in
bytes). The "locked-in-memory size" ulimit (ulimit -l) of the DataNode user also needs to be increased to
match this parameter (see OS Limits). When setting this value, remember that you will need space in memory
for other things as well, such as the DataNode and application JVM heaps and the operating system page
cache.
Optional
The following properties are not required, but may be specified for tuning:
dfs.namenode.path.based.cache.refresh.interval.ms: The NameNode uses this as the amount of
milliseconds between subsequent path cache rescans. This calculates the blocks to cache and each DataNode
containing a replica of the block that should cache it. By default, this parameter is set to 300000, which is
five minutes.
dfs.datanode.fsdatasetcache.max.threads.per.volume: The DataNode uses this as the maximum
number of threads per volume to use for caching new data. By default, this parameter is set to 4.
dfs.cachereport.intervalMsec: The DataNode uses this as the amount of milliseconds between sending
a full report of its cache state to the NameNode. By default, this parameter is set to 10000, which is 10
seconds.
dfs.namenode.path.based.cache.block.map.allocation.percent: The percentage of the Java heap
which we will allocate to the cached blocks map. The cached blocks map is a hash map which uses chained
hashing. Smaller maps may be accessed more slowly if the number of cached blocks is large; larger maps
will consume more memory. By default, this parameter is set to 0.25 percent.

OS Limits
If you get the error Cannot start datanode because the configured max locked memory size... is
more than the datanode's available RLIMIT_MEMLOCK ulimit, that means that the operating system
is imposing a lower limit on the amount of memory that you can lock than what you have configured. To fix this,
you must adjust the ulimit -l value that the DataNode runs with. Usually, this value is configured in
/etc/security/limits.conf. However, it will vary depending on what operating system and distribution you
are using.
You will know that you have correctly configured this value when you can run ulimit -l from the shell and
get back either a higher value than what you have configured with dfs.datanode.max.locked.memory, or the
string unlimited, indicating that there is no limit. Note that it's typical for ulimit -l to output the memory
lock limit in KB, but dfs.datanode.max.locked.memory must be specified in bytes.

Using CDH with Isilon Storage


EMC Isilon is a storage service with a distributed file system that can used in place of HDFS to provide storage
for CDH services.

152 | Cloudera Administration

Managing CDH and Managed Services


Note: This documentation covers only the Cloudera Manager portion of using EMC Isilon storage with
Cloudera Manager. For information about tasks performed on Isilon OneFS, see the information hub
for Cloudera on the EMC Community Network: https://community.emc.com/docs/DOC-39522.
Supported Versions
The following versions of Cloudera and Isilon products are supported:
CDH 5.4.4 or higher, and corresponding level of Impala
Cloudera Manager 5.4.3 or higher
Isilon OneFS 7.2.0.3
Note: Cloudera Navigator is not supported in this release.
Differences between Isilon HDFS and CDH HDFS
The following features of HDFS are not implemented with Isilon OneFS:
HDFS caching
HDFS encryption
HDFS ACLs
Preliminary Steps on the Isilon Service
Before installing a Cloudera Manager cluster to use Isilon storage, perform the following steps on the Isilon
OneFS system. For detailed information on setting up Isilon OneFS for Cloudera Manager, see the Isilon
documentation at https://community.emc.com/docs/DOC-39522.
1. Create an Isilon access zone with HDFS support.
Example:
/ifs/your-access-zone/hdfs

Note: The above is simply an example; the HDFS root directory does not have to begin with ifs
or end with hdfs.
2. Create two directories that will be used by all CDH services:
a. Create a tmp directory in the access zone.
Create supergroup group and hdfs user.
Create a tmp directory and set ownership to hdfs:supergroup, and permissions to 1777.
Example:
cd hdfs_root_directory
isi_run -z zone_id mkdir tmp
isi_run -z zone_id chown hdfs:supergroup tmp
isi_run -z zone_id chmod 1777 tmp

b. Create a user directory in the access zone and set ownership to hdfs:supergroup, and permissions to
755
Example:
cd hdfs_root_directory
isi_run -z zone_id mkdir user
isi_run -z zone_id chown hdfs:supergroup user
isi_run -z zone_id chmod 755 user

3. Create the service-specific users, groups, or directories for each CDH service you plan to use. Create the
directories under the access zone you have created.

Cloudera Administration | 153

Managing CDH and Managed Services


Note: Many of the values provided in the examples below are default values in Cloudera Manager
and must match the Cloudera Manager configuration settings. The format for the examples is:
dir user:group permission . Create the directories below under the access zone you have
created, for example, /ifs/ your-access-zone /hdfs/
ZooKeeper: nothing required.
HBase
Create hbase group with hbase user.
Create the root directory for HBase:
Example:
hdfs_root_directory/hbase hbase:hbase 755

YARN (MR2)
Create mapred group with mapred user.
Create history directory for YARN:
Example:
hdfs_root_directory/user/history mapred:hadoop 777

Create the remote application log directory for YARN:


Example:
hdfs_root_directory/tmp/logs mapred:hadoop 775

Oozie
Create oozie group with oozie user.
Create the user directory for Oozie:
Example:
hdfs_root_directory/user/oozie oozie:oozie 775

Flume
Create flume group with flume user.
Create the user directory for Flume:
Example:
hdfs_root_directory/user/flume flume:flume 775

Hive
Create hive group with hive user.
Create the user directory for Hive:
Example:
hdfs_root_directory/user/hive hive:hive 775

Create the warehouse directory for Hive:


Example:
hdfs_root_directory/user/hive/warehouse hive:hive 1777

154 | Cloudera Administration

Managing CDH and Managed Services


Create a temporary directory for Hive:
Example:
hdfs_root_directory/tmp/hive hive:supergroup 777

Solr
Create solr group with solr user.
Create the data directory for Solr:
Example:
hdfs_root_directory/solr solr:solr 775

Sqoop
Create sqoop group with sqoop2 user.
Create the user directory for Sqoop:
Example:
hdfs_root_directory/user/sqoop2 sqoop2:sqoop 775

Hue
Create hue group with hue user.
Create sample group with sample user.
Spark
Create spark group with spark user.
Create the user directory for Spark:
Example:
hdfs_root_directory/user/spark spark:spark 751

Create application history directory for Spark:


Example:
hdfs_root_directory/user/spark/applicationHistory spark:spark 1777

Once the users, groups, and directories are created in Isilon OneFS, you are ready to install Cloudera Manager.
Installing Cloudera Manager with Isilon
To install Cloudera Manager following the instructions provided in Installing Cloudera Manager, CDH, and Managed
Services.
The simplest installation procedure, suitable for development or proof of concept, is Installation Path A, which
uses embedded databases that are installed as part of the Cloudera Manager installation process.
For production environments, Installation Path B - Manual Installation Using Cloudera Manager Packages
describes configuring external databases for Cloudera Manager and CDH storage needs.
If you choose parcel installation on the Cluster Installation screen, the installation wizard will point to the latest
parcels of CDH available.
On the installation wizard's Cluster Setup page, choose Custom Services, and choose the services you want
installed in the cluster. Be sure to choose Isilon among the selected services, do not select the HDFS service,
and do not check Include Cloudera Navigator at the bottom of the Cluster Setup page. Also, on the Role
Assignments page, be sure to specify the hosts that will serve as gateway roles for the Isilon service. You can
add gateway roles to one, some, or all nodes in the cluster.

Cloudera Administration | 155

Managing CDH and Managed Services


Installing a Secure Cluster with Isilon
To set up a secure cluster with Isilon using Kerberos, perform the following steps:
1. Create an unsecure Cloudera Manager cluster as described above in Installing Cloudera Manager with Isilon
on page 155.
2. Follow the Isilon documentation to enable Kerberos for your access zone:
https://community.emc.com/docs/DOC-39522. This includes adding a Kerberos authentication provider to
your Isilon access zone.
3. Add the following proxy users in Isilon if your Cloudera Manager cluster includes the corresponding CDH
services. The procedure for configuring proxy users is described in the Isilon documentation,
https://community.emc.com/docs/DOC-39522.

proxy user hdfs for hdfs user.


proxy user mapred for mapred user.
proxy user hive for hive user.
proxy user impala for impala user.
proxy user oozie for oozie user
proxy user flume for flume user
proxy user hue for hue user

4. Follow the Cloudera Manager documentation for information on configuring a secure cluster with Kerberos:
Configuring Authentication in Cloudera Manager.
Using Impala with Isilon Storage
You can use Impala to query data files that reside on EMC Isilon storage devices, rather than in HDFS. This
capability allows convenient query access to a storage system where you might already be managing large
volumes of data. The combination of the Impala query engine and Isilon storage is certified on CDH 5.4.4 or
higher.
Because the EMC Isilon storage devices use a global value for the block size rather than a configurable value for
each file, the PARQUET_FILE_SIZE query option has no effect when Impala inserts data into a table or partition
residing on Isilon storage. Use the isi command to set the default block size globally on the Isilon device. For
example, to set the Isilon default block size to 256 MB, the recommended size for Parquet data files for Impala,
issue the following command:
isi hdfs settings modify --default-block-size=256MB

The typical use case for Impala and Isilon together is to use Isilon for the default filesystem, replacing HDFS
entirely. In this configuration, when you create a database, table, or partition, the data always resides on Isilon
storage and you do not need to specify any special LOCATION attribute. If you do specify a LOCATION attribute,
its value refers to a path within the Isilon filesystem. For example:
-- If the default filesystem is Isilon, all Impala data resides there
-- and all Impala databases and tables are located there.
CREATE TABLE t1 (x INT, s STRING);
-- You can specify LOCATION for database, table, or partition,
-- using values from the Isilon filesystem.
CREATE DATABASE d1 LOCATION '/some/path/on/isilon/server/d1.db';
CREATE TABLE d1.t2 (a TINYINT, b BOOLEAN);

Impala can write to, delete, and rename data files and database, table, and partition directories on Isilon storage.
Therefore, Impala statements such as CREATE TABLE, DROP TABLE, CREATE DATABASE, DROP DATABASE, ALTER
TABLE, and INSERT work the same with Isilon storage as with HDFS.
When the Impala spill-to-disk feature is activated by a query that approaches the memory limit, Impala writes
all the temporary data to a local (not Isilon) storage device. Because the I/O bandwidth for the temporary data
depends on the number of local disks, and clusters using Isilon storage might not have as many local disks
attached, pay special attention on Isilon-enabled clusters to any queries that use the spill-to-disk feature. Where
156 | Cloudera Administration

Managing CDH and Managed Services


practical, tune the queries or allocate extra memory for Impala to avoid spilling. Although you can specify an
Isilon storage device as the destination for the temporary data for the spill-to-disk feature, that configuration
is not recommended due to the need to transfer the data both ways using remote I/O.
When tuning Impala queries on HDFS, you typically try to avoid any remote reads. When the data resides on
Isilon storage, all the I/O consists of remote reads. Do not be alarmed when you see non-zero numbers for
remote read measurements in query profile output. The benefit of the Impala and Isilon integration is primarily
convenience of not having to move or copy large volumes of data to HDFS, rather than raw query performance.
You can increase the performance of Impala I/O for Isilon systems by increasing the value for the
num_remote_hdfs_io_threads configuration parameter, in the Cloudera Manager user interface for clusters
using Cloudera Manager, or through the --num_remote_hdfs_io_threads startup option for the impalad
daemon on clusters not using Cloudera Manager.
For information about managing Isilon storage devices through Cloudera Manager, see Using CDH with Isilon
Storage on page 152.

Managing Hive
Use the following procedures to manage HiveServer2 and the Hive metastore. To configure high availability for
the Hive metastore, see Hive Metastore High Availability on page 347.

Heap Size and Garbage Collection for Hive Components


Hive Component Memory Recommendations
HiveServer2 and the Hive metastore require sufficient memory in order to run correctly. The default heap size
of 256 MB for each component is inadequate for production workloads. Consider the following guidelines for
sizing the heap for each component, based upon your cluster size.
Number of Concurrent Connections HiveServer2 Heap Size Minimum
Recommendation

Hive Metastore Heap Size Minimum


Recommendation

Up to 40 concurrent connections
12 GB
(Cloudera recommends splitting
HiveServer2 into multiple instances
and load balancing once you start
allocating >12 GB to HiveServer2.
The objective is to size to reduce
impact of Java garbage collection on
active processing by the service.

12 GB

Up to 20 concurrent connections

6 GB

10 GB

Up to 10 concurrent connections

4 GB

8 GB

Single connection

2 GB

4 GB

Important: These numbers are general guidance only, and may be affected by factors such as number
of columns, partitions, complex joins, and client activity among other things. It is important to review
and refine through testing based on your anticipated deployment to arrive at best values for your
environment.
In addition, the Beeline CLI should use a heap size of at least 2 GB.
The permGenSize should be set to 512M for all.

Cloudera Administration | 157

Managing CDH and Managed Services


Configuring Heap Size and Garbage Collection for Hive Components
To configure the heap size for HiveServer2 and Hive metastore, set the -Xmx parameter in the HADOOP_OPTS
variable to the desired maximum heap size in the hive-env.sh advanced configuration snippet if you use
Cloudera Manager or otherwise edit /etc/hive/hive-env.sh.
To configure the heap size for the Beeline CLI, set the HADOOP_HEAPSIZE environment variable in the hive-env.sh
advanced configuration snippet if you use Cloudera Manager or otherwise edit /etc/hive/hive-env.sh before
starting the Beeline CLI.
The following example shows a configuration with the following settings:
HiveServer2 uses 12 GB heap
Hive metastore uses 12 GB heap
Hive clients use 2 GB heap
The settings to change are in bold. All of these lines are commented out (prefixed with a # character) by default.
Uncomment the lines by removing the # character.
if [ "$SERVICE" = "cli" ]; then
if [ -z "$DEBUG" ]; then
export HADOOP_OPTS="$HADOOP_OPTS -XX:NewRatio=12 -Xmx12288m -Xms10m
-XX:MaxHeapFreeRatio=40 -XX:MinHeapFreeRatio=15 -XX:+useParNewGC -XX:-useGCOverheadLimit"
else
export HADOOP_OPTS="$HADOOP_OPTS -XX:NewRatio=12 -Xmx12288m -Xms10m
-XX:MaxHeapFreeRatio=40 -XX:MinHeapFreeRatio=15 -XX:-useGCOverheadLimit"
fi
fi
export HADOOP_HEAPSIZE=2048

You can choose whether to use the Concurrent Collector or the New Parallel Collector for garbage collection, by
passing -XX:+useParNewGC or -XX:+useConcMarkSweepGC in the HADOOP_OPTS lines above, and you can tune
the garbage collection overhead limit by setting -XX:-useGCOverheadLimit. To enable the garbage collection
overhead limit, remove the setting or change it to -XX:+useGCOverheadLimit.

Configuration for WebHCat


If you want to use WebHCat, you need to set the PYTHON_CMD variable in /etc/default/hive-webhcat-server
after installing Hive; for example:
export PYTHON_CMD=/usr/bin/python

Managing Hive Using Cloudera Manager


There are two Hive service roles:
Hive Metastore Server - manages the metastore process when Hive is configured with a remote metastore.
You are strongly encouraged to read Configuring the Hive Metastore (CDH 4) or Configuring the Hive Metastore
(CDH 5).
HiveServer2 - supports a Thrift API tailored for JDBC and ODBC clients, Kerberos authentication, and multi-client
concurrency. There is also a CLI for HiveServer2 named Beeline. Cloudera recommends that you deploy
HiveServer2 whenever possible. You can use the original HiveServer, and run it concurrently with HiveServer2.
However, Cloudera Manager does not manage HiveServer, so you must configure and manage it outside
Cloudera Manager. See HiveServer2 documentation (CDH 4) or HiveServer2 documentation (CDH 5) for more
information.

158 | Cloudera Administration

Managing CDH and Managed Services


The Hive Metastore Server
Cloudera recommends using a remote Hive metastore, especially for CDH 4.2 or later. Since the remote metastore
is recommended, Cloudera Manager treats the Hive Metastore Server as a required role for all Hive services.
Here are a couple key reasons why the remote metastore setup is advantageous, especially in production settings:
The Hive metastore database password and JDBC drivers dont need to be shared with every Hive client; only
the Hive Metastore Server does. Sharing passwords with many hosts is a security concern.
You can control activity on the Hive metastore database. To stop all activity on the database, just stop the
Hive Metastore Server. This makes it easy to perform tasks such as backup and upgrade, which require all
Hive activity to stop.
Information about the initial configuration of a remote Hive metastore database with Cloudera Manager can be
found at Cloudera Manager and Managed Service Data Stores.
Considerations When Upgrading CDH
Hive has undergone major version changes from CDH 4.0 to 4.1 and between CDH 4.1 and 4.2. (CDH 4.0 had Hive
0.8.0, CDH 4.1 used Hive 0.9.0, and CDH 4.2 or later has 0.10.0). This requires that you manually back up and
upgrade the Hive metastore database when upgrading between major Hive versions.
You should follow the steps in the appropriate in the Cloudera Manager procedure for upgrading CDH to upgrade
the metastore before you restart the Hive service. This applies whether you are upgrading to packages or parcels.
The procedure for upgrading CDH using packages is at Upgrading CDH 4 Using Packages. The procedure for
upgrading with parcels is at Upgrading CDH 4 Using Parcels.
Considerations When Upgrading Cloudera Manager
Cloudera Manager 4.5 added support for Hive, which includes the Hive Metastore Server role type. This role
manages the metastore process when Hive is configured with a remote metastore.
When upgrading from Cloudera Manager versions before 4.5, Cloudera Manager automatically creates new Hive
services to capture the previous implicit Hive dependency from Hue and Impala. Your previous services continue
to function without impact. If Hue was using a Hive metastore backed by a Derby database, the newly created
Hive Metastore Server also uses Derby. Because Derby does not allow concurrent connections, Hue continues
to work, but the new Hive Metastore Server does not run. The failure is harmless (because nothing uses this
new Hive Metastore Server at this point) and intentional, to preserve the set of cluster functionality as it was
before upgrade. Cloudera discourages the use of a Derby-backed Hive metastore due to its limitations and
recommends switching to a different supported database.
Cloudera Manager provides a Hive configuration option to bypass the Hive Metastore Server. When this
configuration is enabled, Hive clients, Hue, and Impala connect directly to the Hive metastore database. Prior to
Cloudera Manager 4.5, Hue and Impala connected directly to the Hive metastore database, so the bypass mode
is enabled by default when upgrading to Cloudera Manager 4.5 or later. This ensures that the upgrade does not
disrupt your existing setup. You should plan to disable the bypass mode, especially when using CDH 4.2 or later.
Using the Hive Metastore Server is the recommended configuration, and the WebHCat Server role requires the
Hive Metastore Server to not be bypassed. To disable bypass mode, see Disabling Bypass Mode on page 159.
Cloudera Manager 4.5 or later also supports HiveServer2 with CDH 4.2. In CDH 4, HiveServer2 is not added by
default, but can be added as a new role under the Hive service (see Role Instances on page 45). In CDH 5,
HiveServer2 is a mandatory role.
Disabling Bypass Mode
Required Role:
In bypass mode Hive clients directly access the metastore database instead of using the Hive Metastore Server
for metastore information.
1. Go to the Hive service.

Cloudera Administration | 159

Managing CDH and Managed Services


2.
3.
4.
5.
6.
7.
8.

Click the Configuration tab.


Select Scope > HIVE service_name (Service-Wide)
Select Category > Advanced.
Deselect the Bypass Hive Metastore Server property.
Click Save Changes to commit the changes.
Re-deploy Hive client configurations.
Restart Hive and any Hue or Impala services configured to use that Hive service.

Using Hive Gateways


Because the Hive service does not have worker roles, another mechanism is needed to enable the automatic
propagation of client configurations to the other hosts in your cluster. Gateway roles fulfill this function. Gateways
in fact aren't really roles and do not have state, but they act as indicators for where client configurations should
be placed. Hive gateways are created by default when the Hive service is added.
Using a Load Balancer with HiveServer2
To configure a load balancer:
1.
2.
3.
4.

Go to the Hive service.


Click the Configuration tab.
Select Scope > HiveServer2.
Enter the hostname and port number of the load balancer in the HiveServer2 Load Balancer property in the
format hostname:port number.
Note:
When you set this property, Cloudera Manager regenerates the keytabs for HiveServer2 roles. The
principal in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this Hive service, it also uses the load balancer to
communicate with Hive.

5. Click Save Changes to commit the changes.

Hive Table Statistics


Required Role:
If your cluster has Impala then you can use the Impala implementation to compute statistics. The Impala
implementation to compute table statistics is available in CDH 5.0.0 or higher and in Impala version 1.2.2 or
higher. The Impala implementation of COMPUTE STATS requires no setup steps and is preferred over the Hive
implementation. See Overview of Table Statistics. If you are running an older version of Impala, you can collect
statistics on a Hive table by running the following command from a Beeline client connected to HiveServer2:
analyze table <table name> compute statistics;
analyze table <table name> compute statistics for columns <all columns of a table>;

Managing User-Defined Functions (UDFs) with HiveServer2


The ADD JAR command does not work with HiveServer2 and the Beeline client when Beeline runs on a different
host. As an alternative to ADD JAR, Hive's auxiliary paths functionality should be used. Perform one of the
following procedures depending on whether you want to create permanent or temporary functions.
User-Defined Functions (UDFs) with HiveServer2 Using Cloudera Manager
Required Role:

160 | Cloudera Administration

Managing CDH and Managed Services


Creating Permanent Functions
1. Copy the JAR file to HDFS and make sure the hive user can access this JAR file.
2. Copy the JAR file to the host on which HiveServer2 is running. Save the JARs to any directory you choose,
give the hive user read, write, and execute access to this directory, and make a note of the path (for example,
/opt/local/hive/lib/).
3. In the Cloudera Manager Admin Console, go to the Hive service.
4. Click the Configuration tab.
5. Expand the Service-Wide > Advanced categories.
6. Configure the Hive Auxiliary JARs Directory property with the HiveServer2 host path from Step 2,
/opt/local/hive/lib/. Setting this property overwrites hive.aux.jars.path, even if this variable has
been previously set in the HiveServer2 advanced configuration snippet.
7. Click Save Changes. The JARs are added to HIVE_AUX_JARS_PATH environment variable.
8. Redeploy the Hive client configuration.
a. In the Cloudera Manager Admin Console, go to the Hive service.
b. From the Actions menu at the top right of the service page, select Deploy Client Configuration.
c. Click Deploy Client Configuration.
9. Restart the Hive service. If the Hive Auxiliary JARs Directory property is configured but the directory does
not exist, HiveServer2 will not start.
10. If Sentry is enabled - Grant privileges on the JAR files to the roles that require access. Login to Beeline as
user hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE

You must also grant privilege to the JAR on HDFS:


GRANT ALL ON URI 'hdfs:///path/to/jar' TO ROLE EXAMPLE_ROLE

11. Run the CREATE FUNCTION command and point to the JAR file location in HDFS. For example:
CREATE FUNCTION addfunc AS 'com.example.hiveserver2.udf.add' USING JAR
'hdfs:///path/to/jar'

Creating Temporary Functions


1. Copy the JAR file to the host on which HiveServer2 is running. Save the JARs to any directory you choose,
give the hive user read, write, and execute access to this directory, and make a note of the path (for example,
/opt/local/hive/lib/).
2. In the Cloudera Manager Admin Console, go to the Hive service.
3. Click the Configuration tab.
4. Expand the Service-Wide > Advanced categories.
5. Configure the Hive Auxiliary JARs Directory property with the HiveServer2 host path from Step 2,
/opt/local/hive/lib/. Setting this property overwrites hive.aux.jars.path, even if this variable has
been previously set in the HiveServer2 advanced configuration snippet.
6. Click Save Changes. The JARs are added to HIVE_AUX_JARS_PATH environment variable.
7. Redeploy the Hive client configuration.
a. In the Cloudera Manager Admin Console, go to the Hive service.
b. From the Actions menu at the top right of the service page, select Deploy Client Configuration.
c. Click Deploy Client Configuration.
8. Restart the Hive service. If the Hive Auxiliary JARs Directory property is configured but the directory does
not exist, HiveServer2 will not start.

Cloudera Administration | 161

Managing CDH and Managed Services


9. If Sentry is enabled - Grant privileges on the JAR files to the roles that require access. Login to Beeline as
user hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE

You must also grant privilege to the JAR on HDFS:


GRANT ALL ON URI 'hdfs:///path/to/jar' TO ROLE EXAMPLE_ROLE

10. Run the CREATE TEMPORARY FUNCTION command. For example:


CREATE TEMPORARY FUNCTION addfunc AS 'com.example.hiveserver2.udf.add'

User-Defined Functions (UDFs) with HiveServer2 Using the Command Line


The following sections describe how to create permanent and temporary functions using the command line.

Creating Permanent Functions


1. Copy the JAR file to HDFS and make sure the hive user can access this jar file.
2. On the Beeline client machine, in /etc/hive/conf/hive-site.xml, set the hive.aux.jars.path property
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
hive.aux.jars.path=file:///opt/local/hive/lib/my.jar

3. Copy the JAR file (and its dependent libraries) to the host running HiveServer2/Impala. Make sure the hive
user has read, write, and execute access to these files on the HiveServer2/Impala host.
4. On the HiveServer2/Impala host, open /etc/default/hive-server2 and set the AUX_CLASSPATH variable
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
AUX_CLASSPATH=/opt/local/hive/lib/my.jar

5. Restart HiveServer2.
6. If Sentry is enabled - Grant privileges on the JAR files to the roles that require access. Login to Beeline as
user hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE

You must also grant privilege to the JAR on HDFS:


GRANT ALL ON URI 'hdfs:///path/to/jar' TO ROLE EXAMPLE_ROLE

If you are using Sentry policy files, you can grant the URI privilege as follows:
udf_r = server=server1->uri=file:///opt/local/hive/lib
udf_r = server=server1->uri=hdfs:///path/to/jar

7. Run the CREATE FUNCTION command and point to the JAR from Hive:
CREATE FUNCTION addfunc AS 'com.example.hiveserver2.udf.add' USING JAR
'hdfs:///path/to/jar'

Creating Temporary Functions


1. On the Beeline client machine, in /etc/hive/conf/hive-site.xml, set the hive.aux.jars.path property
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
hive.aux.jars.path=file:///opt/local/hive/lib/my.jar

162 | Cloudera Administration

Managing CDH and Managed Services


2. Copy the JAR file (and its dependent libraries) to the host running HiveServer2/Impala. Make sure the hive
user has read, write, and execute access to these files on the HiveServer2/Impala host.
3. On the HiveServer2/Impala host, open /etc/default/hive-server2 and set the AUX_CLASSPATH variable
to a comma-separated list of the fully-qualified paths to the JAR file and any dependent libraries.
AUX_CLASSPATH=/opt/local/hive/lib/my.jar

4. If Sentry is enabled - Grant privileges on the local JAR files to the roles that require access. Login to Beeline
as user hive and use the Hive SQL GRANT statement to do so. For example:
GRANT ALL ON URI 'file:///opt/local/hive/lib/my.jar' TO ROLE EXAMPLE_ROLE

If you are using Sentry policy files, you can grant the URI privilege as follows:
udf_r = server=server1->uri=file:///opt/local/hive/lib

5. Restart HiveServer2.
6. Run the CREATE TEMPORARY FUNCTION command and point to the JAR from Hive:
CREATE TEMPORARY FUNCTION addfunc AS 'com.example.hiveserver2.udf.add'

Hive on Spark
This section explains how to set up Hive on Spark. It assumes that your cluster is managed by Cloudera Manager.
Important: Hive on Spark is included in CDH 5.4 and higher but is not currently supported nor
recommended for production use. To try this feature, use it in a test environment until Cloudera
resolves currently existing issues and limitations to make it ready for production use.
Configuring Hive on Spark
Important: Hive on Spark is included in CDH 5.4 and higher but is not currently supported nor
recommended for production use. To try this feature, use it in a test environment until Cloudera
resolves currently existing issues and limitations to make it ready for production use.
This topic explains the configuration properties you set up to run Hive on Spark.
Note: We recommend that you use HiveServer2 with Beeline. The following content, except for
Configuring Hive on Spark for Hive CLI on page 165, is based on this assumption.
Installation Considerations
For Hive to work on Spark, you must deploy Spark gateway roles on the same machine that hosts HiveServer2.
Otherwise, Hive on Spark cannot read from Spark configurations and cannot submit Spark jobs. For more
information about gateway roles, see Managing Roles on page 44.
After installation, run the following command in Hive so that Hive will use Spark as the back-end engine for all
subsequent queries.
set hive.execution.engine=spark;

Enabling Hive on Spark


By default, Hive on Spark is not enabled. To enable Hive on Spark, perform the following steps in Cloudera
Manager.
1. Go to the Hive service.
Cloudera Administration | 163

Managing CDH and Managed Services


2.
3.
4.
5.
6.

Click the Configuration tab.


Enter Enable Hive on Spark in the Search field.
Check the box for Enable Hive on Spark (Unsupported).
Locate the Spark On YARN Service and click SPARK_ON_YARN.
Click Save Changes to commit the changes.

Configuration Properties
Property

Description

hive.stats.collect.rawdatasize

Hive on Spark uses statistics to determine the


threshold for converting common join to map join.
There are two types of statistics about the size of data:
totalSize: The approximate size of data on the
disk
rawDataSize: The approximate size of data in
memory
When both metrics are available, Hive chooses
rawDataSize.
Default: True

hive.auto.convert.join.noconditionaltask.size The threshold for the sum of all the small table size
(by default, rawDataSize), for map join conversion. You

can increase the value if you want better performance


by converting more common joins to map joins.
However, if you set this value too high, tasks may fail
because too much memory is being used by data from
small tables.
Default: 20MB
Configuring Hive
For improved performance, Cloudera recommends that you configure the following additional properties for
Hive. In Cloudera Manager, set these properties in the advanced configuration snippet for HiveServer2.
hive.stats.fetch.column.stats=true
hive.optimize.index.filter=true
Configuring Executor Memory Size
For general Spark configuration recommendations, see Configuring Spark on YARN Applications on page 239.
Executor memory size can have a number of effects on Hive. Increasing executor memory increases the number
of queries for which Hive can enable mapjoin optimization. However, if there's too much executor memory, it
takes longer to perform garbage collection. Also, some experiments shows that HDFS doesnt handle concurrent
writers well, so it may face a race condition if there are too many executor cores.
Cloudera recommends that you set the value for spark.executor.cores to 5, 6, or 7, depending on what the
host is divisible by. For example, if yarn.nodemanager.resource.cpu-vcores is 19, then you would set the
value to 6. Executors must have the same number of cores. If you set the value to 5, each executor only gets
three cores, with four left unused. If you set the value to 7, only two executors are used, and five cores are
unused. If the number of cores is 20, set the value to 5 so that each executor gets four cores, and no cores are
unused.
Cloudera also recommends the following:

164 | Cloudera Administration

Managing CDH and Managed Services


Compute a memory size equal to yarn.nodemanager.resource.memory-mb * (spark.executor.cores
/ yarn.nodemanager.resource.cpu-vcores) and then split that between spark.executor.memory and
spark.yarn.executor.memoryOverhead.
spark.yarn.executor.memoryOverhead is 15-20% of the total memory size.
Troubleshooting Hive on Spark
Important: Hive on Spark is included in CDH 5.4 and higher but is not currently supported nor
recommended for production use. To try this feature, use it in a test environment until Cloudera
resolves currently existing issues and limitations to make it ready for production use.
Problem: Delayed result from the first query after starting a new Hive on Spark session
The first query after starting a new Hive on Spark session might be delayed due to the start-up time for
the Spark on YARN cluster. The query waits for YARN containers to initialize. Subsequent queries will be
faster.
Problem: Exception Error: org.apache.thrift.transport.TTransportException (state=08S01,code=0)
and HiveServer2 is down
HiveServer2 memory is set too small. For more information, see STDOUT for HiveServer2. To fix this issue:
1. In Cloudera Manager, go to HIVE.
2. Click Configuration.
3. Search for Java Heap Size of HiveServer2 in Bytes, and change it to be a larger value. Cloudera
recommends a minimum value of 256 MB.
4. Restart HiveServer2.
Problem: Out-of-memory error
You might get an out-of-memory error similar to the following:
15/03/19 03:43:17 WARN channel.DefaultChannelPipeline: An exception was thrown
by a user handler while handling an exception event ([id: 0x9e79a9b1,
/10.20.118.103:45603 => /10.20.120.116:39110] EXCEPTION:
java.lang.OutOfMemoryError: Java heap space)
java.lang.OutOfMemoryError: Java heap space

This error indicates that the Spark driver does not have enough off-heap memory. Increase the off-heap
memory by setting spark.yarn.driver.memoryOverhead or spark.driver.memory.
Problem: Hive on Spark does not work with HBase
Hive on Spark with HBase is not supported. If you use HBase, use Hive on MapReduce instead of Hive
on Spark.
Problem: Spark applications stay alive forever and occupy cluster resources
This can occur if there are multiple concurrent Hive sessions. To manually terminate the Spark applications:
1. Find the YARN application IDs for the applications by going to Cloudera Manager and clicking Yarn >
ResourceManager > ResourceManager Web UI.
2. Log in to the YARN ResourceManager host.
3. Open a terminal and run:
yarn application -kill <applicationID>
applicationID is each YARN application ID you found in step 1.

Configuring Hive on Spark for Hive CLI


You no longer need to configure Hive on Spark for the Hive command line interface (CLI). This feature is now
configured automatically.

Cloudera Administration | 165

Managing CDH and Managed Services


Important: Hive on Spark is included in CDH 5.4 and higher but is not currently supported nor
recommended for production use. To try this feature, use it in a test environment until Cloudera
resolves currently existing issues and limitations to make it ready for production use.

Managing Hue
Hue is a set of web UIs that enable you to interact with a CDH cluster. This section describes tasks for managing
Hue.

Adding a Hue Service and Role Instance


Adding the Hue Service
Required Role:
After initial installation, you can use the Add a Service wizard to add and configure a new Hue service instance.
1.

On the Home > Status tab, click


types display.
2. Select Hue.
3. Click Continue.

to the right of the cluster name and select Add a Service. A list of service

A page displays where you can specify the dependencies for the Hue service.
4. Select the row with the Hue dependencies required for your cluster. For more information, see Hue
Dependencies.
5. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
6. Click Continue.
Cloudera Manager starts the Hue service.
7. Click Continue.
8. Click Finish.
9. If your cluster uses Kerberos, Cloudera Manager will automatically add a Hue Kerberos Ticket Renewer role
to each host where you assigned the Hue Server role instance. Also see, Enable Hue to Work with Hadoop
Security using Cloudera Manager.

166 | Cloudera Administration

Managing CDH and Managed Services


Adding a Hue Role Instance
Required Role:
1.
2.
3.
4.

In Cloudera Manager Administration Console, navigate to the Hue service.


Click the Instances tab.
Click the Add Role Instances button.
Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. If your cluster uses Kerberos, you must add the Hue Kerberos Ticket Renewer role to each host where you
assigned the Hue Server role instance. Cloudera Manager will throw a validation error if the new Hue Server
role does not have a colocated KT Renewer role. Also see, Enable Hue to Work with Hadoop Security using
Cloudera Manager.
6. Click Continue.

Hue and High Availability


If your cluster has HDFS high availability enabled, you must configure the Hue HDFS Web Interface Role property
to use HttpFS. See Configuring Hue to Work with HDFS HA on page 310 for detailed instructions.
To configure the Hue service itself for high availability, see Hue High Availability on page 344.

Managing Hue Analytics Data Collection


Required Role:
Hue tracks anonymized pages and application versions to collect information used to compare each application's
usage levels. The data collected does not include hostnames or IDs; For example, the data has the format
/2.3.0/pig, /2.5.0/beeswax/execute. You can restrict data collection as follows:
1.
2.
3.
4.
5.

Go to the Hue service.


Click the Configuration tab.
Select Scope > Hue.
Locate the Enable Usage Data Collection property or search for it by typing its name in the Search box.
Deselect the Enable Usage Data Collection checkbox.

Cloudera Administration | 167

Managing CDH and Managed Services


If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
6. Click Save Changes to commit the changes.
7. Restart the Hue service.

Enabling Hue Applications Using Cloudera Manager


Required Role:
Most Hue applications are configured by default, based on the services you have installed. Cloudera Manager
selects the service instance that Hue depends on. If you have more than one service, you may want to verify or
change the service dependency for Hue. Also, if you add a service such as Sqoop 2 or Oozie after you have set
up Hue, you need to set the dependency because it is not done automatically. To add a dependency:
1.
2.
3.
4.
5.
6.
7.

Go to the Hue service.


Click the Configuration tab.
Select Scope > Hue (Service-Wide).
Select Category > Main.
Select each service name Service property to set the dependency. Select none to remove the dependency.
Click Save Changes to commit the changes.
Restart the Hue service.

Enabling the Sqoop 2 Application


If you are upgrading Cloudera Manager from a release 4.6 or lower, you need to set the Hue dependency to
enable the Sqoop 2 application.
Enabling the HBase Browser Application with doAs Impersonation
Required Role:
The Hue HBase application communicates through a proxy server called the HBase Thrift Server, which then
forwards commands to HBase. Because Hue stands between the Thrift server and the actual user, all HBase
operations appear to come from the hue user and not the actual user. To secure these interactions, you must
do the following:
Ensure that users logged into Hue perform operations with their own privileges, and not those of the
impersonating hue user.
Once Hue can impersonate other users, ensure that only the Hue server can send commands to the HBase
Thrift server. To ensure this, use Kerberos to authenticate the hue user to the HBase Thrift server.
To enable the HBase browser application:
1. Add the HBase Thrift Server role.
2. If you have a Kerberos-enabled cluster, enable impersonation by configuring the following HBase properties:
a.
b.
c.
d.
e.

Select the HBase service.


Click the Configuration tab.
Select Scope > Service-Wide.
Select Category > Security.
For the HBase Thrift Authentication property, make sure it is set to one of the following values:
auth-conf: authentication, integrity and confidentiality checking
auth-int: authentication and integrity checking
auth: authentication only

f. Select Category > Main.

168 | Cloudera Administration

Managing CDH and Managed Services


g. Check the Enable HBase Thrift Http Server and Enable HBase Thrift Proxy Users properties checkboxes.
h. Click Save Changes to commit the changes.
3. Enable TLS/SSL for the HBase Thrift Server.
4. Configure Hue to point to the Thrift Server and to a valid HBase configuration directory:
a.
b.
c.
d.
e.

Select the Hue service.


Click the Configuration tab.
Select Scope > All.
Select Category > Main.
For the HBase Service property, make sure it is set to the HBase service for which you enabled the Thrift
Server role (if you have more than one HBase service instance).
f. In the HBase Thrift Server property, click the edit field and select the Thrift Server role for Hue to use.
g. Select Category > Advanced.
h. Locate the Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini property
and add the following property:
[hbase]
hbase_conf_dir=/etc/hbase/conf

i. Click Save Changes to commit the changes.


Enabling the Solr Search Application
To use the Solr Search application with Hue, you must update the URL for the Solr Server in the Hue Server
advanced configuration snippet. In addition, if you are using parcels with CDH 4.3, you must register the
"hue-search" application manually, or access will fail. See Deploying Solr with Hue on page 219 for detailed
instructions.

Using an External Database for Hue


Cloudera strongly recommends an external database for clusters with multiple Hue users, especially clusters
in a production environment. The default database, SQLite, works best with a single user and a small dataset.
For supported databases, see:
CDH 4 supported databases
CDH 5 supported databases
Using an External Database for Hue Using Cloudera Manager
Required Role:
The Hue server requires an SQL database to store small amounts of data such as user account information, job
submissions, and Hive queries. Hue supports a lightweight embedded (SQLite) database and several types of
external databases. This page explains how to configure Hue with a selection of Supported Databases.
Important: Cloudera strongly recommends an external database for clusters with multiple Hue users.
To configure Hue with any of the supported external databases, the high-level steps are:
1.
2.
3.
4.
5.

Stop Hue service.


Backup default SQLite database (if applicable).
Install database software and dependencies.
Create and configure database and load data.
Start Hue service.

See the tasks on this page for details. If you don't need to migrate a SQLite database, you can skip the steps on
dumping the database and editing the JSON objects.
Cloudera Administration | 169

Managing CDH and Managed Services


Configuring the Hue Server to Store Data in MariaDB
For information about installing and configuring a MariaDB database , see MariaDB Database.
1.
2.
3.
4.

In the Cloudera Manager Admin Console, go to the Hue service status page.
Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
Note the host to which the dump was written under Step in the Dump Database Command window. You can
also find it by selecting Commands > Recent Commands > Dump Database.
5. Open a terminal window for the host and navigate to the dump file in /tmp/hue_database_dump.json.
6. Remove all JSON objects with useradmin.userprofile in the model field, for example:
{
"pk": 14,
"model": "useradmin.userprofile",
"fields":
{ "creation_method": "EXTERNAL", "user": 14, "home_directory": "/user/tuser2" }
},

7. Set strict mode in /etc/my.cnf and restart MySQL:


[mysqld]
sql_mode=STRICT_ALL_TABLES

8. Create a new database and grant privileges to a Hue user to manage this database. For example:
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by 'secretpassword';
Query OK, 0 rows affected (0.00 sec)

9. In the Cloudera Manager Admin Console, click the Hue service.


10. Click the Configuration tab.
11. Select Scope > All.
12. Select Category > Database.
13. Specify the settings for Hue Database Type, Hue Database Hostname, Hue Database Port, Hue Database
Username, Hue Database Password, and Hue Database Name. For example, for a MySQL database on the
local host, you might use the following values:

Hue Database Type = mysql


Hue Database Hostname = host
Hue Database Port = 3306
Hue Database Username = hue
Hue Database Password = secretpassword
Hue Database Name = hue

14. Optionally restore the Hue data to the new database:


a. Select Actions > Synchronize Database.
b. Determine the foreign key ID.
$ mysql -uhue -psecretpassword
mysql > SHOW CREATE TABLE auth_permission;

c. (InnoDB only) Drop the foreign key that you retrieved in the previous step.
mysql > ALTER TABLE auth_permission DROP FOREIGN KEY
content_type_id_refs_id_XXXXXX;

170 | Cloudera Administration

Managing CDH and Managed Services


d. Delete the rows in the django_content_type table.
mysql > DELETE FROM hue.django_content_type;

e. In Hue service instance page, click Actions > Load Database. Confirm you want to load the database by
clicking Load Database.
f. (InnoDB only) Add back the foreign key.
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (content_type_id) REFERENCES
django_content_type (id);

15. Start the Hue service.

Configuring the Hue Server to Store Data in MySQL


Note: Cloudera recommends InnoDB over MyISAM as the Hue MySQL engine. On CDH 5, Hue requires
InnoDB.
For information about installing and configuring a MySQL database , see MySQL Database.
1.
2.
3.
4.

In the Cloudera Manager Admin Console, go to the Hue service status page.
Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
Note the host to which the dump was written under Step in the Dump Database Command window. You can
also find it by selecting Commands > Recent Commands > Dump Database.
5. Open a terminal window for the host and navigate to the dump file in /tmp/hue_database_dump.json.
6. Remove all JSON objects with useradmin.userprofile in the model field, for example:
{
"pk": 14,
"model": "useradmin.userprofile",
"fields":
{ "creation_method": "EXTERNAL", "user": 14, "home_directory": "/user/tuser2" }
},

7. Set strict mode in /etc/my.cnf and restart MySQL:


[mysqld]
sql_mode=STRICT_ALL_TABLES

8. Create a new database and grant privileges to a Hue user to manage this database. For example:
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by 'secretpassword';
Query OK, 0 rows affected (0.00 sec)

9. In the Cloudera Manager Admin Console, click the Hue service.


10. Click the Configuration tab.
11. Select Scope > All.
12. Select Category > Database.
13. Specify the settings for Hue Database Type, Hue Database Hostname, Hue Database Port, Hue Database
Username, Hue Database Password, and Hue Database Name. For example, for a MySQL database on the
local host, you might use the following values:

Hue Database Type = mysql


Hue Database Hostname = host
Hue Database Port = 3306
Hue Database Username = hue
Cloudera Administration | 171

Managing CDH and Managed Services


Hue Database Password = secretpassword
Hue Database Name = hue
14. Optionally restore the Hue data to the new database:
a. Select Actions > Synchronize Database.
b. Determine the foreign key ID.
$ mysql -uhue -psecretpassword
mysql > SHOW CREATE TABLE auth_permission;

c. (InnoDB only) Drop the foreign key that you retrieved in the previous step.
mysql > ALTER TABLE auth_permission DROP FOREIGN KEY
content_type_id_refs_id_XXXXXX;

d. Delete the rows in the django_content_type table.


mysql > DELETE FROM hue.django_content_type;

e. In Hue service instance page, click Actions > Load Database. Confirm you want to load the database by
clicking Load Database.
f. (InnoDB only) Add back the foreign key.
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (content_type_id) REFERENCES
django_content_type (id);

15. Start the Hue service.

Configuring the Hue Server to Store Data in PostgreSQL


For information about installing and configuring an external PostgreSQL database , see External PostgreSQL
Database.
1.
2.
3.
4.

In the Cloudera Manager Admin Console, go to the Hue service status page.
Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
Note the host to which the dump was written under Step in the Dump Database Command window. You can
also find it by selecting Commands > Recent Commands > Dump Database.
5. Open a terminal window for the host and navigate to the dump file in /tmp/hue_database_dump.json.
6. Remove all JSON objects with useradmin.userprofile in the model field, for example:
{
"pk": 14,
"model": "useradmin.userprofile",
"fields":
{ "creation_method": "EXTERNAL", "user": 14, "home_directory": "/user/tuser2" }
},

7. Install the PostgreSQL server.


RHEL
$ sudo yum install postgresql-server

SLES
$ sudo zypper install postgresql-server

172 | Cloudera Administration

Managing CDH and Managed Services


Ubuntu or Debian
$ sudo apt-get install postgresql

8. Initialize the data directories.


$ service postgresql initdb

9. Configure client authentication.


a. Edit /var/lib/pgsql/data/pg_hba.conf.
b. Set the authentication methods for local to trust and for host to password and add the following line
at the end.
host hue hue 0.0.0.0/0 md5

10. Start the PostgreSQL server.


$ su - postgres
# /usr/bin/postgres -D /var/lib/pgsql/data > logfile 2>&1 &

11. Configure PostgreSQL to listen on all network interfaces.


a. Edit /var/lib/pgsql/data/postgresql.conf and set list_addresses.
listen_addresses = 0.0.0.0

# Listen on all addresses

12. Create the hue database and grant privileges to a hue user to manage the database.
# psql -U postgres
postgres=# create database hue;
postgres=# \c hue;
You are now connected to database 'hue'.
postgres=# create user hue with password 'secretpassword';
postgres=# grant all privileges on database hue to hue;
postgres=# \q

13. Restart the PostgreSQL server.


$ sudo service postgresql restart

14. Verify connectivity.


psql h localhost U hue d hue
Password for user hue: secretpassword

15. Configure the PostgreSQL server to start at boot.


RHEL
$ sudo /sbin/chkconfig postgresql on
$ sudo /sbin/chkconfig --list postgresql
postgresql
0:off
1:off
2:on

3:on

4:on

5:on

6:off

SLES
$ sudo chkconfig --add postgresql

Cloudera Administration | 173

Managing CDH and Managed Services


Ubuntu or Debian
$ sudo chkconfig postgresql on

16. Configure the Hue database:


a.
b.
c.
d.
e.

In the Cloudera Manager Admin Console, click the HUE service.


Click the Configuration tab.
Select Scope > Hue Server.
Select Category > Advanced.
Set Hue Server Advanced Configuration Snippet (Safety Valve) for hue_safety_valve_server.ini with the
following:
[desktop]
[[database]]
engine=postgresql_psycopg2
name=hue
host=localhost
port=5432
user=hue
password=secretpassword

Note: If you set Hue Database Hostname, Hue Database Port, Hue Database Username, and
Hue Database Password at the service-level, under Service-Wide > Database, you can omit
those properties from the server-lever configuration above and avoid storing the Hue password
as plain text. In either case, set engine and name in the server-level safety-valve.
f. Click Save Changes.
17. Optionally restore the Hue data to the new database:
a. Select Actions > Synchronize Database.
b. Determine the foreign key ID.
bash# su postgres
$ psql h localhost U hue d hue
postgres=# \d auth_permission;

c. Drop the foreign key that you retrieved in the previous step.
postgres=# ALTER TABLE auth_permission DROP CONSTRAINT
content_type_id_refs_id_XXXXXX;

d. Delete the rows in the django_content_type table.


postgres=# TRUNCATE django_content_type CASCADE;

e. In Hue service instance page, Actions > Load Database. Confirm you want to load the database by clicking
Load Database.
f. Add back the foreign key you dropped.
bash# su postgres
$ psql h localhost U hue d hue
postgres=# ALTER TABLE auth_permission ADD CONSTRAINT
content_type_id_refs_id_XXXXXX FOREIGN KEY (content_type_id) REFERENCES
django_content_type(id) DEFERRABLE INITIALLY DEFERRED;

18. Start the Hue service.

174 | Cloudera Administration

Managing CDH and Managed Services


Configuring the Hue Server to Store Data in Oracle (Parcel Installation)
Use the following instructions to configure the Hue Server with an Oracle database if you are working on a
parcel-based deployment. If you are using packages, see Configuring the Hue Server to Store Data in Oracle
(Package Installation) on page 176.
For information about installing and configuring an Oracle database , see Oracle Database.
Important: Configure the database for character set AL32UTF8 and national character set UTF8.
1. Install the required packages.
RHEL
$ sudo yum install gcc python-devel python-pip python-setuptools libaio

SLES
$ sudo zypper install gcc python-devel python-pip python-setuptools libaio

Ubuntu or Debian
$ sudo apt-get install gcc python-devel python-pip python-setuptools libaio1

2. Add http://tiny.cloudera.com/hue-oracle-client-db to the Cloudera Manager remote parcel repository URL


list and download, distribute, and activate the parcel.
3. For CDH versions lower than 5.3, install the Python Oracle library:
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ HUE_HOME/build/env/bin/pip install cx_Oracle

4. For CDH versions lower than 5.3, upgrade django south:


$ HUE_HOME/build/env/bin/pip install south --upgrade

5. In the Cloudera Manager Admin Console, go to the Hue service status page.
6. Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
7. Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
8. Click the Configuration tab.
9. Select Scope > All.
10. Select Category > Advanced.
11. Set the Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini property.
Note: If you set Hue Database Hostname, Hue Database Port, Hue Database Username, and Hue
Database Password at the service-level, under Service-Wide > Database, you can omit those
properties from the server-lever configuration above and avoid storing the Hue password as plain
text. In either case, set engine and name in the server-level safety-valve.
Add the following options (and modify accordingly for your setup):
[desktop]
[[database]]
host=localhost
port=1521

Cloudera Administration | 175

Managing CDH and Managed Services


engine=oracle
user=hue
password=secretpassword
name=<SID of the Oracle database, for example, 'XE'>

For CDH 5.1 and higher you can use an Oracle service name. To use the Oracle service name instead of the
SID, use the following configuration instead:
port=0
engine=oracle
user=hue
password=secretpassword
name=oracle.example.com:1521/orcl.example.com

The directive port=0 allows Hue to use a service name. The name string is the connect string, including
hostname, port, and service name.
To add support for a multithreaded environment, set the threaded option to true under the
[desktop]>[[database]] section.
options={"threaded":true}

12. Grant required permissions to the hue user in Oracle:


grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant

alter any index to hue;


alter any table to hue;
alter database link to hue;
create any index to hue;
create any sequence to hue;
create database link to hue;
create session to hue;
create table to hue;
drop any sequence to hue;
select any dictionary to hue;
drop any table to hue;
create procedure to hue;
create trigger to hue;
execute on sys.dbms_lob to hue;

13. Navigate to the Hue Server instance in Cloudera Manager and select Actions > Synchronize Database.
14. Ensure you are connected to Oracle as the hue user, then run the following command to delete all data from
Oracle tables:
> set pagesize 100;
> SELECT 'DELETE FROM ' || table_name || ';' FROM user_tables;

15. Run the statements generated in the preceding step.


16. Commit your changes.
commit;

17. Load the data that you dumped. Navigate to the Hue Server instance and select Actions > Load Database.
This step is not necessary if you have a fresh Hue install with no data or if you dont want to save the Hue
data.
18. Start the Hue service.

Configuring the Hue Server to Store Data in Oracle (Package Installation)


If you have a parcel-based environment, see Configuring the Hue Server to Store Data in Oracle (Parcel Installation)
on page 175.
Important: Configure the database for character set AL32UTF8 and national character set UTF8.

176 | Cloudera Administration

Managing CDH and Managed Services


1. Download the Oracle libraries at Instant Client for Linux x86-64 Version 11.1.0.7.0, Basic and SDK (with
headers) zip files to the same directory.
2. Unzip the Oracle client zip files.
3. Set environment variables to reference the libraries.
$ export ORACLE_HOME=oracle_download_directory
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME

4. Create a symbolic link for the shared object:


$ cd $ORACLE_HOME
$ ln -sf libclntsh.so.11.1 libclntsh.so

5. Install the required packages.


RHEL
$ sudo yum install gcc python-devel python-pip python-setuptools libaio

SLES
$ sudo zypper install gcc python-devel python-pip python-setuptools libaio

Ubuntu or Debian
$ sudo apt-get install gcc python-devel python-pip python-setuptools libaio1

6. For CDH versions lower than 5.3, install the Python Oracle library:
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ HUE_HOME/build/env/bin/pip install cx_Oracle

7. For CDH versions lower than 5.3, upgrade django south:


$ HUE_HOME/build/env/bin/pip install south --upgrade

8. In the Cloudera Manager Admin Console, go to the Hue service status page.
9. Select Actions > Stop. Confirm you want to stop the service by clicking Stop.
10. Select Actions > Dump Database. Confirm you want to dump the database by clicking Dump Database.
11. Click the Configuration tab.
12. Select Scope > All.
13. Select Category > Advanced.
14. Set the Hue Service Environment Advanced Configuration Snippet (Safety Valve) property to
ORACLE_HOME=oracle_download_directory
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:oracle_download_directory

15. Set the Hue Service Advanced Configuration Snippet (Safety Valve) for hue_safety_valve.ini property.
Note: If you set Hue Database Hostname, Hue Database Port, Hue Database Username, and Hue
Database Password at the service-level, under Service-Wide > Database, you can omit those
properties from the server-lever configuration above and avoid storing the Hue password as plain
text. In either case, set engine and name in the server-level safety-valve.

Cloudera Administration | 177

Managing CDH and Managed Services


Add the following options (and modify accordingly for your setup):
[desktop]
[[database]]
host=localhost
port=1521
engine=oracle
user=hue
password=secretpassword
name=<SID of the Oracle database, for example, 'XE'>

For CDH 5.1 and higher you can use an Oracle service name. To use the Oracle service name instead of the
SID, use the following configuration instead:
port=0
engine=oracle
user=hue
password=secretpassword
name=oracle.example.com:1521/orcl.example.com

The directive port=0 allows Hue to use a service name. The name string is the connect string, including
hostname, port, and service name.
To add support for a multithreaded environment, set the threaded option to true under the
[desktop]>[[database]] section.
options={"threaded":true}

16. Grant required permissions to the hue user in Oracle:


grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant

alter any index to hue;


alter any table to hue;
alter database link to hue;
create any index to hue;
create any sequence to hue;
create database link to hue;
create session to hue;
create table to hue;
drop any sequence to hue;
select any dictionary to hue;
drop any table to hue;
create procedure to hue;
create trigger to hue;
execute on sys.dbms_lob to hue;

17. Navigate to the Hue Server instance in Cloudera Manager and select Actions > Synchronize Database.
18. Ensure you are connected to Oracle as the hue user, then run the following command to delete all data from
Oracle tables:
> set pagesize 100;
> SELECT 'DELETE FROM ' || table_name || ';' FROM user_tables;

19. Run the statements generated in the preceding step.


20. Commit your changes.
commit;

21. Load the data that you dumped. Navigate to the Hue Server instance and select Actions > Load Database.
This step is not necessary if you have a fresh Hue install with no data or if you dont want to save the Hue
data.
22. Start the Hue service.

178 | Cloudera Administration

Managing CDH and Managed Services


Using an External Database for Hue Using the Command Line
The Hue server requires a SQL database to store small amounts of data such as user account information, job
submissions, and Hive queries. SQLite is the default embedded database. Hue also supports several types of
external databases. This page explains how to configure Hue with a selection of external Supported Databases.
Important: Cloudera strongly recommends an external database for clusters with multiple Hue users.
To configure Hue with any of the supported external databases, the high-level steps are:
1.
2.
3.
4.
5.

Stop Hue service.


Backup default SQLite database.
Install database software and dependencies.
Create and configure database and load data.
Start Hue service.

See the tasks on this page for details. If you don't need to migrate a SQLite database, you can skip the steps on
dumping the database and editing the JSON objects.

Prerequisites
Before using an external database with Hue, install all of the support libraries required by your operating system.
See Development Preferences in the Hue documentation for the full list.

Embedded Database
By default, Hue is configured to use the embedded database, SQLite, and should require no configuration or
management by the administrator.
Inspecting the Embedded Hue Database
The default SQLite database used by Hue is located in /var/lib/hue/desktop.db. You can inspect this database
from the command line using the sqlite3 program. For example:
# sqlite3 /var/lib/hue/desktop.db
SQLite version 3.6.22
Enter ".help" for instructions
Enter SQL statements terminated with a ";"
sqlite> select username from auth_user;
admin
test
sample
sqlite>

Important: It is strongly recommended that you avoid making any modifications to the database
directly using sqlite3, though sqlite3 is useful for management or troubleshooting.
Backing up the Embedded Hue Database
If you use the default embedded SQLite database, copy the desktop.db file to another node for backup. It is
recommended that you back it up on a regular schedule, and also that you back it up before any upgrade to a
new version of Hue.

External Database
Cloudera strongly recommends an external database for clusters with multiple Hue users, especially clusters
in a production environment. The default database, SQLite, cannot support large data migrations. Hue supports
MariaDB, MySQL, PostgreSQL, and Oracle. See Supported Databases for the supported versions.
In the instructions that follow, dumping the database and editing the JSON objects is only necessary if you have
data in SQLite that you need to migrate. If you don't need to migrate data from SQLite, you can skip those steps.

Cloudera Administration | 179

Managing CDH and Managed Services


Configuring the Hue Server to Store Data in MariaDB
Note: Cloudera recommends InnoDB over MyISAM as the Hue MySQL engine. On CDH 5, Hue requires
InnoDB.
1. Shut down the Hue server if it is running.
2. Open <some-temporary-file>.json and remove all JSON objects with useradmin.userprofile in the
model field. Here are some examples of JSON objects that should be deleted.
{
"pk": 1,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1,
"home_directory": "/user/alice"
}
},
{
"pk": 2,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1100714,
"home_directory": "/user/bob"
}
},
.....

3. Start the Hue server.


4. Install the MariaDB client developer package.
OS

Command

RHEL

$ sudo yum install mariadb-devel

SLES

$ sudo zypper install mariadb-devel

Ubuntu or Debian

$ sudo apt-get install libmariadbclient-dev

5. Install the MariaDB connector.


OS

Command

RHEL

$ sudo yum install mariadb-connector-java

SLES

$ sudo zypper install mariadb-connector-java

Ubuntu or Debian

$ sudo apt-get install libmariadb-java

6. Install and start MariaDB.


OS

Command

RHEL

$ sudo yum install mariadb-server

SLES

$ sudo zypper install

Ubuntu or Debian

$ sudo apt-get install mariadb-server

180 | Cloudera Administration

mariadb-server
$ sudo zypper install libmariadblclient18

Managing CDH and Managed Services


7. Change the /etc/my.cnf file as follows:
[mysqld]
datadir=/var/lib/mysql
socket=/var/lib/mysql/mysql.sock
bind-address=<ip-address>
default-storage-engine=InnoDB
sql_mode=STRICT_ALL_TABLES

8. Start the MariaDB daemon.


$ sudo service mariadb start

9. Configure MariaDB to use a strong password. In the following procedure, your current root password is
blank. Press the Enter key when you're prompted for the root password.
$ sudo /usr/bin/mysql_secure_installation
[...]
Enter current password for root (enter for none):
OK, successfully used password, moving on...
[...]
Set root password? [Y/n] y
New password:
Re-enter new password:
Remove anonymous users? [Y/n] Y
[...]
Disallow root login remotely? [Y/n] N
[...]
Remove test database and access to it [Y/n] Y
[...]
Reload privilege tables now? [Y/n] Y
All done!

10. Configure MariaDB to start at boot.


OS

Command

RHEL

$ sudo /sbin/chkconfig mariadb on


$ sudo /sbin/chkconfig --list mariadb
mysqld
0:off
1:off
2:on
6:off

SLES

$ sudo chkconfig --add mariadb

Ubuntu or Debian

$ sudo chkconfig mariadb on

3:on

4:on

5:on

11. Create the Hue database and grant privileges to a hue user to manage the database.
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by '<secretpassword>';
Query OK, 0 rows affected (0.00 sec)

12. Open the Hue configuration file in a text editor.


13. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=3306
engine=mysql
user=hue
password=<secretpassword>
name=hue

Cloudera Administration | 181

Managing CDH and Managed Services


14. As the hue user, load the existing data and create the necessary database tables using syncdb and migrate
commands. When running these commands, Hue will try to access a logs directory, located at
/opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that is the case, first create the
logs directory and give the hue user and group ownership of the directory.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo -u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
$ sudo -u hue <HUE_HOME>/build/env/bin/hue migrate
$ mysql -u hue -p <secretpassword>
mysql > SHOW CREATE TABLE auth_permission;

15. (InnoDB only) Drop the foreign key.


mysql > ALTER TABLE auth_permission DROP FOREIGN KEY content_type_id_refs_id_XXXXXX;

16. Delete the rows in the django_content_type table.


mysql > DELETE FROM hue.django_content_type;

17. Load the data.


$ <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json

18. (InnoDB only) Add the foreign key.


$ mysql -u hue -p <secretpassword>
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (`content_type_id`) REFERENCES
`django_content_type` (`id`);

Configuring the Hue Server to Store Data in MySQL


Note: Cloudera recommends InnoDB over MyISAM as the Hue MySQL engine. On CDH 5, Hue requires
InnoDB.
1. Shut down the Hue server if it is running.
2. Dump the existing database data to a text file. Note that using the .json extension is required.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo -u hue <HUE_HOME>/build/env/bin/hue dumpdata > <some-temporary-file>.json

3. Open <some-temporary-file>.json and remove all JSON objects with useradmin.userprofile in the
model field. Here are some examples of JSON objects that should be deleted.
{
"pk": 1,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1,

182 | Cloudera Administration

Managing CDH and Managed Services


"home_directory":

"/user/alice"

}
},
{
"pk": 2,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1100714,
"home_directory": "/user/bob"
}
},
.....

4. Start the Hue server.


5. Install the MySQL client developer package.
OS

Command

RHEL

$ sudo yum install mysql-devel

SLES

$ sudo zypper install mysql-devel

Ubuntu or Debian

$ sudo apt-get install libmysqlclient-dev

6. Install the MySQL connector.


OS

Command

RHEL

$ sudo yum install mysql-connector-java

SLES

$ sudo zypper install mysql-connector-java

Ubuntu or Debian

$ sudo apt-get install libmysql-java

7. Install and start MySQL.


OS

Command

RHEL

$ sudo yum install mysql-server

SLES

$ sudo zypper install mysql


$ sudo zypper install libmysqlclient_r15

Ubuntu or Debian

$ sudo apt-get install mysql-server

8. Change the /etc/my.cnf file as follows:


[mysqld]
datadir=/var/lib/mysql
socket=/var/lib/mysql/mysql.sock
bind-address=<ip-address>
default-storage-engine=InnoDB
sql_mode=STRICT_ALL_TABLES

9. Start the mysql daemon.


OS

Command

RHEL

$ sudo service mysqld start

SLES and Ubuntu or


Debian

$ sudo service mysql start

Cloudera Administration | 183

Managing CDH and Managed Services


10. Configure MySQL to use a strong password. In the following procedure, your current root password is blank.
Press the Enter key when you're prompted for the root password.
$ sudo /usr/bin/mysql_secure_installation
[...]
Enter current password for root (enter for none):
OK, successfully used password, moving on...
[...]
Set root password? [Y/n] y
New password:
Re-enter new password:
Remove anonymous users? [Y/n] Y
[...]
Disallow root login remotely? [Y/n] N
[...]
Remove test database and access to it [Y/n] Y
[...]
Reload privilege tables now? [Y/n] Y
All done!

11. Configure MySQL to start at boot.


OS

Command

RHEL

$ sudo /sbin/chkconfig mysqld on


$ sudo /sbin/chkconfig --list mysqld
mysqld
0:off
1:off
2:on
6:off

SLES

$ sudo chkconfig --add mysql

Ubuntu or Debian

$ sudo chkconfig mysql on

3:on

4:on

5:on

12. Create the Hue database and grant privileges to a hue user to manage the database.
mysql> create database hue;
Query OK, 1 row affected (0.01 sec)
mysql> grant all on hue.* to 'hue'@'localhost' identified by '<secretpassword>';
Query OK, 0 rows affected (0.00 sec)

13. Open the Hue configuration file in a text editor.


14. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=3306
engine=mysql
user=hue
password=<secretpassword>
name=hue

15. As the hue user, load the existing data and create the necessary database tables using syncdb and migrate
commands. When running these commands, Hue will try to access a logs directory, located at
/opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that is the case, first create the
logs directory and give the hue user and group ownership of the directory.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
$ sudo -u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput

184 | Cloudera Administration

Managing CDH and Managed Services


$ sudo -u hue <HUE_HOME>/build/env/bin/hue migrate
$ mysql -u hue -p <secretpassword>
mysql > SHOW CREATE TABLE auth_permission;

16. (InnoDB only) Drop the foreign key.


mysql > ALTER TABLE auth_permission DROP FOREIGN KEY content_type_id_refs_id_XXXXXX;

17. Delete the rows in the django_content_type table.


mysql > DELETE FROM hue.django_content_type;

18. Load the data.


$ <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json

19. (InnoDB only) Add the foreign key.


$ mysql -u hue -p <secretpassword>
mysql > ALTER TABLE auth_permission ADD FOREIGN KEY (`content_type_id`) REFERENCES
`django_content_type` (`id`);

Configuring the Hue Server to Store Data in PostgreSQL


Warning: Hue requires PostgreSQL 8.4 or newer.
1. Shut down the Hue server if it is running.
2. Dump the existing database data to a text file. Note that using the .json extension is required.
Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ sudo -u hue <HUE_HOME>/build/env/bin/hue dumpdata > <some-temporary-file>.json

3. Open <some-temporary-file>.json and remove all JSON objects with useradmin.userprofile in the
model field. Here are some examples of JSON objects that should be deleted.
{
"pk": 1,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1,
"home_directory": "/user/alice"
}
},
{
"pk": 2,
"model": "useradmin.userprofile",
"fields": {
"creation_method": "HUE",
"user": 1100714,
"home_directory": "/user/bob"
}
},
.....

4. Install required packages.

Cloudera Administration | 185

Managing CDH and Managed Services


OS

Command

RHEL

$ sudo yum install postgresql-devel gcc python-devel

SLES

$ sudo zypper install postgresql-devel gcc python-devel

Ubuntu or Debian

$ sudo apt-get install postgresql-devel gcc python-devel

5. Install the module that provides the connector to PostgreSQL.


sudo -u hue <HUE_HOME>/build/env/bin/pip install setuptools
sudo -u hue <HUE_HOME>/build/env/bin/pip install psycopg2

6. Install the PostgreSQL server.


OS

Command

RHEL

$ sudo yum install postgresql-server

SLES

$ sudo zypper install postgresql-server

Ubuntu or Debian

$ sudo apt-get install postgresql

7. Initialize the data directories:


$ service postgresql initdb

8. Configure client authentication.


a. Edit /var/lib/pgsql/data/pg_hba.conf.
b. Set the authentication methods for local to trust and for host to password and add the following line
at the end.
host hue hue 0.0.0.0/0 md5

9. Start the PostgreSQL server.


$ su - postgres
# /usr/bin/postgres -D /var/lib/pgsql/data > logfile 2>&1 &

10. Configure PostgreSQL to listen on all network interfaces.


Edit /var/lib/pgsql/data/postgresql.conf and set list_addresses:
listen_addresses = '0.0.0.0'

# Listen on all addresses

11. Create the hue database and grant privileges to a hue user to manage the database.
# psql -U postgres
postgres=# create database hue;
postgres=# \c hue;
You are now connected to database 'hue'.
postgres=# create user hue with password '<secretpassword>';
postgres=# grant all privileges on database hue to hue;
postgres=# \q

12. Restart the PostgreSQL server.


$ sudo service postgresql restart

186 | Cloudera Administration

Managing CDH and Managed Services


13. Verify connectivity.
psql -h localhost -U hue -d hue
Password for user hue: <secretpassword>

14. Configure the PostgreSQL server to start at boot.


OS

Command

RHEL

$ sudo /sbin/chkconfig postgresql on


$ sudo /sbin/chkconfig --list postgresql
postgresql
0:off
1:off
2:on
6:off

SLES

$ sudo chkconfig --add postgresql

Ubuntu or Debian

$ sudo chkconfig postgresql on

3:on

4:on

5:on

15. Open the Hue configuration file in a text editor.


16. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=5432
engine=postgresql_psycopg2
user=hue
password=<secretpassword>
name=hue

17. As the hue user, configure Hue to load the existing data and create the necessary database tables. You will
need to run both the migrate and syncdb commands. When running these commands, Hue will try to access
a logs directory, located at /opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that
is the case, first create the logs directory and give the hue user and group ownership of the directory.
$
$
$
$

sudo
sudo
sudo
sudo

mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
-u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
-u hue <HUE_HOME>/build/env/bin/hue migrate

18. Determine the foreign key ID.


bash# su - postgres
$ psql -h localhost -U hue -d hue
postgres=# \d auth_permission;

19. Drop the foreign key that you retrieved in the previous step.
postgres=# ALTER TABLE auth_permission DROP CONSTRAINT
content_type_id_refs_id_<XXXXXX>;

20. Delete the rows in the django_content_type table.


postgres=# TRUNCATE django_content_type CASCADE;

21. Load the data.


$ sudo -u hue <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json

Cloudera Administration | 187

Managing CDH and Managed Services


22. Add back the foreign key you dropped.
bash# su - postgres
$ psql -h localhost -U hue -d hue
postgres=# ALTER TABLE auth_permission ADD CONSTRAINT
content_type_id_refs_id_<XXXXXX> FOREIGN KEY (content_type_id) REFERENCES
django_content_type(id) DEFERRABLE INITIALLY DEFERRED;

Configuring the Hue Server to Store Data in Oracle


Important: Configure the database for character set AL32UTF8 and national character set UTF8.
1. Ensure Python 2.6 or newer is installed on the server Hue is running on.
2. Download the Oracle client libraries at Instant Client for Linux x86-64 Version 11.1.0.7.0, Basic and SDK (with
headers) zip files to the same directory.
3. Unzip the zip files.
4. Set environment variables to reference the libraries.
$ export ORACLE_HOME=<download directory>
$ export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$ORACLE_HOME

5. Create a symbolic link for the shared object:


$ cd $ORACLE_HOME
$ ln -sf libclntsh.so.11.1 libclntsh.so

6. Get a data dump by executing:


Note: HUE_HOME is a reference to the location of your Hue installation. For package installs, this
is usually /usr/lib/hue; for parcel installs, this is usually, /opt/cloudera/parcels/<parcel
version>/lib/hue/.
$ <HUE_HOME>/build/env/bin/hue dumpdata > <some-temporary-file>.json --indent 2

7. Edit the Hue configuration file hue.ini. Directly below the [[database]] section under the [desktop] line,
add the following options (and modify accordingly for your setup):
host=localhost
port=1521
engine=oracle
user=hue
password=<secretpassword>
name=<SID of the Oracle database, for example, 'XE'>

To use the Oracle service name instead of the SID, use the following configuration instead:
port=0
engine=oracle
user=hue
password=password
name=oracle.example.com:1521/orcl.example.com

The directive port=0 allows Hue to use a service name. The name string is the connect string, including
hostname, port, and service name.
To add support for a multithreaded environment, set the threaded option to true under the
[desktop]>[[database]] section.
options={'threaded':true}

188 | Cloudera Administration

Managing CDH and Managed Services


8. Grant required permissions to the Hue user in Oracle:
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant
grant

alter any index to hue;


alter any table to hue;
alter database link to hue;
create any index to hue;
create any sequence to hue;
create database link to hue;
create session to hue;
create table to hue;
drop any sequence to hue;
select any dictionary to hue;
drop any table to hue;
create procedure to hue;
create trigger to hue;
execute on sys.dbms_lob to hue;

9. As the hue user, configure Hue to load the existing data and create the necessary database tables. You will
need to run both the syncdb and migrate commands. When running these commands, Hue will try to access
a logs directory, located at /opt/cloudera/parcels/CDH/lib/hue/logs, which might be missing. If that
is the case, first create the logs directory and give the hue user and group ownership of the directory.
$
$
$
$

sudo
sudo
sudo
sudo

mkdir /opt/cloudera/parcels/CDH/lib/hue/logs
chown hue:hue /opt/cloudera/parcels/CDH/lib/hue/logs
-u hue <HUE_HOME>/build/env/bin/hue syncdb --noinput
-u hue <HUE_HOME>/build/env/bin/hue migrate

10. Ensure you are connected to Oracle as the hue user, then run the following command to delete all data from
Oracle tables:
SELECT 'DELETE FROM ' || '.' || table_name || ';' FROM user_tables;

11. Run the statements generated in the preceding step.


12. Commit your changes.
commit;

13. Load the data.


$ sudo -u hue <HUE_HOME>/build/env/bin/hue loaddata <some-temporary-file>.json

Managing Impala
This section explains how to configure Impala to accept connections from applications that use popular
programming APIs:
Post-Installation Configuration for Impala on page 193
Configuring Impala to Work with ODBC on page 196
Configuring Impala to Work with JDBC on page 198
This type of configuration is especially useful when using Impala in combination with Business Intelligence tools,
which use these standard interfaces to query different kinds of database and Big Data systems.
You can also configure these other aspects of Impala:
Overview of Impala Security
Modifying Impala Startup Options

The Impala Service


You can install Cloudera Impala through the Cloudera Manager installation wizard, using either parcels or
packages, and have the service created and started as part of the Installation wizard. See Installing Impala.
Cloudera Administration | 189

Managing CDH and Managed Services


If you elect not to include the Impala service using the Installation wizard, you can use the Add Service wizard
to perform the installation. The wizard will automatically configure and start the dependent services and the
Impala service. See Adding a Service on page 36 for instructions.
For further information on the Impala service, see:
CDH 4 - Cloudera Impala Documentation for CDH 4
CDH 5 - Impala Guide
For information on features that support Impala resource management see Impala Resource Management on
page 265.
Configuring the Impala Service
There are several types of configuration settings you may need to apply, depending on your situation.
Running Impala with CDH 4.1
If you are running CDH 4.1, and the Bypass Hive Metastore Server option is enabled, do the following:
1.
2.
3.
4.

Go to the Impala service.


Click the Configuration tab.
Select Impala Daemon > Advanced.
Add the following to the Impala Daemon Hive Advanced Configuration Snippet (Safety Valve) property,
replacing hive_metastore_server_host with the name of your Hive Metastore Server host:
<property>
<name>hive.metastore.local</name>
<value>false</value>
</property>
<property>
<name>hive.metastore.uris</name>
<value>thrift://hive_metastore_server_host:9083</value>
</property>

5. Click Save Changes to commit the changes.


6. Restart the Impala service.
Configuring Table Statistics
Configuring table statistics is highly recommended when using Impala. It allows Impala to make optimizations
that can result in significant (over 10x) performance improvement for some joins. If these are not available,
Impala will still function, but at lower performance.
The Impala implementation to compute table statistics is available in CDH 5.0.0 or higher and in Impala version
1.2.2 or higher. The Impala implementation of COMPUTE STATS requires no setup steps and is preferred over
the Hive implementation. See Overview of Table Statistics. If you are running an older version of Impala,follow
the procedure in Hive Table Statistics on page 160.
Using a Load Balancer with Impala
To configure a load balancer:
1.
2.
3.
4.
5.

Go to the Impala service.


Click the Configuration tab.
Select Scope > Impala Daemon
Select Category > All
Enter the hostname and port number of the load balancer in the Impala Daemons Load Balancer property
in the format hostname:port number.

190 | Cloudera Administration

Managing CDH and Managed Services


Note:
When you set this property, Cloudera Manager regenerates the keytabs for Impala Daemon roles.
The principal in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this Impala service, it also uses the load balancer to
communicate with Impala.
6. Click Save Changes to commit the changes.
Impala Web Servers
Enabling and Disabling Access to Impala Web Servers
Each of the Impala-related daemons includes a built-in web server that lets an administrator diagnose issues
with each daemon on a particular host, or perform other administrative actions such as cancelling a running
query. By default, these web servers are enabled. You might turn them off in a high-security configuration where
it is not appropriate for users to have access to this kind of monitoring information through a web interface. (To
leave the web servers enabled but control who can access their web pages, consult the Configuring Secure
Access for Impala Web Servers later in this section.)
Impala Daemon
1.
2.
3.
4.
5.
6.
7.

Go to the Impala service.


Click the Configuration tab.
Select Scope > Impala Daemon
Select Category > Ports and Addresses.
Select or deselect Enable Impala Daemon Web Server.
Click Save Changes to commit the changes.
Restart the Impala service.

Impala StateStore
1.
2.
3.
4.
5.
6.
7.

Go to the Impala service.


Click the Configuration tab.
Select Scope > Impala StateStore.
Select Category > All
Select or deselect Enable StateStore Web Server.
Click Save Changes to commit the changes.
Restart the Impala service.

Impala Catalog Server


1.
2.
3.
4.
5.
6.
7.

Go to the Impala service.


Click the Configuration tab.
Select Scope > Impala Catalog Server.
Select Category > All
Check or uncheck Enable Catalog Server Web Server.
Click Save Changes to commit the changes.
Restart the Impala service.

Opening Impala Web Server UIs


Impala StateStore
1. Go to the Impala service.
2. Select Web UI > Impala StateStore Web UI.

Cloudera Administration | 191

Managing CDH and Managed Services


Impala Daemon
1.
2.
3.
4.

Go the to Impala service.


Click the Instances tab.
Click an Impala Daemon instance.
Click Impala Daemon Web UI.

Impala Catalog Server


1. Go to the Impala service.
2. Select Web UI > Impala Catalog Web UI.
Impala Llama ApplicationMaster
1.
2.
3.
4.

Go to the Impala service.


Click the Instances tab.
Click a Impala Llama ApplicationMaster instance.
Click Llama Web UI.

Configuring Secure Access for Impala Web Servers


Cloudera Manager supports two methods of authentication for secure access to the Impala Catalog Server,
Daemon, and StateStoreweb servers: password-based authentication and TLS/SSL certificate authentication.
Both of these can be configured through properties of the Impala Catalog Server, Daemon, and StateStore.
Authentication for the three types of daemons can be configured independently.
Configuring Password Authentication
1. Go to the Impala service.
2. Click the Configuration tab.
3. Search for "password" using the Search box within the Configuration page. This should display the
password-related properties (Username and Password properties) for the Impala Catalog Server, Daemon,
and StateStore. If there are multiple role groups configured for Impala Daemon instances, the search should
display all of them.
4. Enter a username and password into these fields.
5. Click Save Changes to commit the changes.
6. Restart the Impala service.
Now when you access the Web UI for the Impala Catalog Server, Daemon, and StateStore, you are asked to log
in before access is granted.
Configuring TLS/SSL Certificate Authentication
1. Create or obtain an TLS/SSL certificate.
2. Place the certificate, in .pem format, on the hosts where the Impala Catalog Server and StateStore are running,
and on each host where an Impala Daemon is running. It can be placed in any location (path) you choose. If
all the Impala Daemons are members of the same role group, then the .pem file must have the same path
on every host.
3. Go to the Impala service page.
4. Click the Configuration tab.
5. Search for "certificate" using the Search box within the Configuration page. This should display the certificate
file location properties for the Impala Catalog Server, Daemon, and StateStore. If there are multiple role groups
configured for Impala Daemon instances, the search should display all of them.
6. In the property fields, enter the full path name to the certificate file.
7. Click Save Changes to commit the changes.
8. Restart the Impala service.

192 | Cloudera Administration

Managing CDH and Managed Services


When you access the Web UI for the Impala Catalog Server, Daemon, and StateStore, https will be used.

Post-Installation Configuration for Impala


This section describes the mandatory and recommended configuration settings for Cloudera Impala. If Impala
is installed using Cloudera Manager, some of these configurations are completed automatically; you must still
configure short-circuit reads manually. If you installed Impala without Cloudera Manager, or if you want to
customize your environment, consider making the changes described in this topic.
In some cases, depending on the level of Impala, CDH, and Cloudera Manager, you might need to add particular
component configuration details in one of the free-form fields on the Impala configuration pages within Cloudera
Manager. In Cloudera Manager 4, these fields are labelled Safety Valve; in Cloudera Manager 5, they are called
Advanced Configuration Snippet.
You must enable short-circuit reads, whether or not Impala was installed through Cloudera Manager. This
setting goes in the Impala configuration settings, not the Hadoop-wide settings.
If you installed Impala in an environment that is not managed by Cloudera Manager, you must enable block
location tracking, and you can optionally enable native checksumming for optimal performance.
If you deployed Impala using Cloudera Manager see Testing Impala Performance to confirm proper
configuration.
Mandatory: Short-Circuit Reads
Enabling short-circuit reads allows Impala to read local data directly from the file system. This removes the
need to communicate through the DataNodes, improving performance. This setting also minimizes the number
of additional copies of data. Short-circuit reads requires libhadoop.so (the Hadoop Native Library) to be
accessible to both the server and the client. libhadoop.so is not available if you have installed from a tarball.
You must install from an .rpm, .deb, or parcel to use short-circuit local reads.
Note: If you use Cloudera Manager, you can enable short-circuit reads through a checkbox in the user
interface and that setting takes effect for Impala as well.
Cloudera strongly recommends using Impala with CDH 4.2 or higher, ideally the latest 4.x release. Impala does
support short-circuit reads with CDH 4.1, but for best performance, upgrade to CDH 4.3 or higher. The process
of configuring short-circuit reads varies according to which version of CDH you are using. Choose the procedure
that is appropriate for your environment.
To configure DataNodes for short-circuit reads with CDH 4.2 or higher:
1. Copy the client core-site.xml and hdfs-site.xml configuration files from the Hadoop configuration
directory to the Impala configuration directory. The default Impala configuration location is /etc/impala/conf.
2. On all Impala nodes, configure the following properties in Impala's copy of hdfs-site.xml as shown:
<property>
<name>dfs.client.read.shortcircuit</name>
<value>true</value>
</property>
<property>
<name>dfs.domain.socket.path</name>
<value>/var/run/hdfs-sockets/dn</value>
</property>
<property>
<name>dfs.client.file-block-storage-locations.timeout.millis</name>
<value>10000</value>
</property>

3. If /var/run/hadoop-hdfs/ is group-writable, make sure its group is root.

Cloudera Administration | 193

Managing CDH and Managed Services


Note: If you are also going to enable block location tracking, you can skip copying configuration
files and restarting DataNodes and go straight to Optional: Block Location Tracking. Configuring
short-circuit reads and block location tracking require the same process of copying files and
restarting services, so you can complete that process once when you have completed all
configuration changes. Whether you copy files and restart services now or during configuring block
location tracking, short-circuit reads are not enabled until you complete those final steps.
4. After applying these changes, restart all DataNodes.
To configure DataNodes for short-circuit reads with CDH 4.1:
Note: Cloudera strongly recommends using Impala with CDH 4.2 or higher, ideally the latest 4.x
release. Impala does support short-circuit reads with CDH 4.1, but for best performance, upgrade to
CDH 4.3 or higher. The process of configuring short-circuit reads varies according to which version of
CDH you are using. Choose the procedure that is appropriate for your environment.
1. Enable short-circuit reads by adding settings to the Impala core-site.xml file.
If you installed Impala using Cloudera Manager, short-circuit reads should be properly configured, but
you can review the configuration by checking the contents of the core-site.xml file, which is installed
at /etc/impala/conf by default.
If you installed using packages, instead of using Cloudera Manager, create the core-site.xml file. This
can be easily done by copying the core-site.xml client configuration file from another machine that is
running Hadoop services. This file must be copied to the Impala configuration directory. The Impala
configuration directory is set by the IMPALA_CONF_DIR environment variable and is by default
/etc/impala/conf. To confirm the Impala configuration directory, check the IMPALA_CONF_DIR
environment variable value.
Note: If the Impala configuration directory does not exist, create it and then add
the core-site.xml file.
Add the following to the core-site.xml file:
<property>
<name>dfs.client.read.shortcircuit</name>
<value>true</value>
</property>

Note: For an installation managed by Cloudera Manager, specify these settings in the Impala
dialogs, in the options field for HDFS. In Cloudera Manager 4, these fields are labelled Safety Valve;
in Cloudera Manager 5, they are called Advanced Configuration Snippet.
2. For each DataNode, enable access by adding the following to the hdfs-site.xml file:
<property>
<name>dfs.client.use.legacy.blockreader.local</name>
<value>true</value>
</property>
<property>
<name>dfs.datanode.data.dir.perm</name>
<value>750</value>
</property>
<property>
<name>dfs.block.local-path-access.user</name>
<value>impala</value>
</property>

194 | Cloudera Administration

Managing CDH and Managed Services


<property>
<name>dfs.client.file-block-storage-locations.timeout.millis</name>
<value>10000</value>
</property>

Note: In the preceding example, the dfs.block.local-path-access.user is the user running


the impalad process. By default, that account is impala.
3. Use usermod to add users requiring local block access to the appropriate HDFS group. For example, if you
assigned impala to the dfs.block.local-path-access.user property, you would add impala to the
hadoop HDFS group:
$ usermod -a -G hadoop impala

Note: The default HDFS group is hadoop, but it is possible to have an environment configured to
use an alternate group. To find the configured HDFS group name using the Cloudera Manager
Admin Console:
1.
2.
3.
4.
5.

Go to the HDFS service.


Click the Configuration tab.
Click Scope > HDFS service name (Service-Wide).
Click Category > Advanced.
The Shared Hadoop Group Name property contains the group name.

Note: If you are going to enable block location tracking, you can skip copying configuration files
and restarting DataNodes and go straight to Mandatory: Block Location Tracking on page 195.
Configuring short-circuit reads and block location tracking require the same process of copying
files and restarting services, so you can complete that process once when you have completed all
configuration changes. Whether you copy files and restart services now or during configuring block
location tracking, short-circuit reads are not enabled until you complete those final steps.
4. Copy the client core-site.xml and hdfs-site.xml configuration files from the Hadoop configuration
directory to the Impala configuration directory. The default Impala configuration location is /etc/impala/conf.
5. After applying these changes, restart all DataNodes.
Mandatory: Block Location Tracking
Enabling block location metadata allows Impala to know which disk data blocks are located on, allowing better
utilization of the underlying disks. Impala will not start unless this setting is enabled.
To enable block location tracking:
1. For each DataNode, adding the following to the hdfs-site.xml file:
<property>
<name>dfs.datanode.hdfs-blocks-metadata.enabled</name>
<value>true</value>
</property>

2. Copy the client core-site.xml and hdfs-site.xml configuration files from the Hadoop configuration
directory to the Impala configuration directory. The default Impala configuration location is /etc/impala/conf.
3. After applying these changes, restart all DataNodes.

Cloudera Administration | 195

Managing CDH and Managed Services


Optional: Native Checksumming
Enabling native checksumming causes Impala to use an optimized native library for computing checksums, if
that library is available.
To enable native checksumming:
If you installed CDH from packages, the native checksumming library is installed and setup correctly. In such a
case, no additional steps are required. Conversely, if you installed by other means, such as with tarballs, native
checksumming may not be available due to missing shared objects. Finding the message "Unable to load
native-hadoop library for your platform... using builtin-java classes where applicable"
in the Impala logs indicates native checksumming may be unavailable. To enable native checksumming, you
must build and install libhadoop.so (the Hadoop Native Library).

Configuring Impala to Work with ODBC


Third-party products can be designed to integrate with Impala using ODBC. For the best experience, ensure any
third-party product you intend to use is supported. Verifying support includes checking that the versions of
Impala, ODBC, the operating system, and the third-party product have all been approved for use together. Before
configuring your systems to use ODBC, download a connector. You may need to sign in and accept license
agreements before accessing the pages required for downloading ODBC connectors.
Downloading the ODBC Driver
Important: As of late 2015, most business intelligence applications are certified with the 2.x ODBC
drivers. Although the instructions on this page cover both the 2.x and 1.x drivers, expect to use the
2.x drivers exclusively for most ODBC applications connecting to Impala.
See the downloads page for a matrix of the certified driver version for different products. See the documentation
page for installation instructions.
Configuring the ODBC Port
Versions 2.5 and 2.0 of the Cloudera ODBC Connector, currently certified for some but not all BI applications, use
the HiveServer2 protocol, corresponding to Impala port 21050. Impala supports Kerberos authentication with
all the supported versions of the driver, and requires ODBC 2.05.13 for Impala or higher for LDAP
username/password authentication.
Version 1.x of the Cloudera ODBC Connector uses the original HiveServer1 protocol, corresponding to Impala
port 21000.
Example of Setting Up an ODBC Application for Impala
To illustrate the outline of the setup process, here is a transcript of a session to set up all required drivers and
a business intelligence application that uses the ODBC driver, under Mac OS X. Each .dmg file runs a GUI-based
installer, first for the underlying IODBC driver needed for non-Windows systems, then for the Cloudera ODBC
Connector, and finally for the BI tool itself.
$ ls -1
Cloudera-ODBC-Driver-for-Impala-Install-Guide.pdf
BI_Tool_Installer.dmg
iodbc-sdk-3.52.7-macosx-10.5.dmg
ClouderaImpalaODBC.dmg
$ open iodbc-sdk-3.52.7-macosx-10.dmg
Install the IODBC driver using its installer
$ open ClouderaImpalaODBC.dmg
Install the Cloudera ODBC Connector using its installer
$ installer_dir=$(pwd)
$ cd /opt/cloudera/impalaodbc
$ ls -1
Cloudera ODBC Driver for Impala Install Guide.pdf
Readme.txt
Setup
lib
ErrorMessages

196 | Cloudera Administration

Managing CDH and Managed Services


Release Notes.txt
Tools
$ cd Setup
$ ls
odbc.ini odbcinst.ini
$ cp odbc.ini ~/.odbc.ini
$ vi ~/.odbc.ini
$ cat ~/.odbc.ini
[ODBC]
# Specify any global ODBC configuration here such as ODBC tracing.
[ODBC Data Sources]
Sample Cloudera Impala DSN=Cloudera ODBC Driver for Impala
[Sample Cloudera Impala DSN]
# Description: DSN Description.
# This key is not necessary and is only to give a description of the data source.
Description=Cloudera ODBC Driver for Impala DSN
# Driver: The location where the ODBC driver is installed to.
Driver=/opt/cloudera/impalaodbc/lib/universal/libclouderaimpalaodbc.dylib
# The DriverUnicodeEncoding setting is only used for SimbaDM
# When set to 1, SimbaDM runs in UTF-16 mode.
# When set to 2, SimbaDM runs in UTF-8 mode.
#DriverUnicodeEncoding=2
# Values for HOST, PORT, KrbFQDN, and KrbServiceName should be set here.
# They can also be specified on the connection string.
HOST=hostname.sample.example.com
PORT=21050
Schema=default
# The authentication mechanism.
# 0 - No authentication (NOSASL)
# 1 - Kerberos authentication (SASL)
# 2 - Username authentication (SASL)
# 3 - Username/password authentication (SASL)
# 4 - Username/password authentication with SSL (SASL)
# 5 - No authentication with SSL (NOSASL)
# 6 - Username/password authentication (NOSASL)
AuthMech=0
# Kerberos related settings.
KrbFQDN=
KrbRealm=
KrbServiceName=
# Username/password authentication with SSL settings.
UID=
PWD
CAIssuedCertNamesMismatch=1
TrustedCerts=/opt/cloudera/impalaodbc/lib/universal/cacerts.pem
# Specify the proxy user ID to use.
#DelegationUID=
# General settings
TSaslTransportBufSize=1000
RowsFetchedPerBlock=10000
SocketTimeout=0
StringColumnLength=32767
UseNativeQuery=0
$ pwd
/opt/cloudera/impalaodbc/Setup
$ cd $installer_dir
$ open BI_Tool_Installer.dmg
Install the BI tool using its installer
$ ls /Applications | grep BI_Tool
BI_Tool.app
$ open -a BI_Tool.app
In the BI tool, connect to a data source using port 21050

Cloudera Administration | 197

Managing CDH and Managed Services


Notes about JDBC and ODBC Interaction with Impala SQL Features
Most Impala SQL features work equivalently through the impala-shell interpreter of the JDBC or ODBC APIs.
The following are some exceptions to keep in mind when switching between the interactive shell and applications
using the APIs:
Note: If your JDBC or ODBC application connects to Impala through a load balancer such as haproxy,
be cautious about reusing the connections. If the load balancer has set up connection timeout values,
either check the connection frequently so that it never sits idle longer than the load balancer timeout
value, or check the connection validity before using it and create a new one if the connection has been
closed.
The Impala complex types (STRUCT, ARRAY, or MAP) are available in CDH 5.5 / Impala 2.3 and higher. To use
these types with JDBC requires version 2.5.28 or higher of the Cloudera JDBC Connector for Impala. To use
these types with ODBC requires version 2.5.30 or higher of the Cloudera ODBC Connector for Impala. Consider
upgrading all JDBC and ODBC drivers at the same time you upgrade from CDH 5.5 or higher.
Although the result sets from queries involving complex types consist of all scalar values, the queries involve
join notation and column references that might not be understood by a particular JDBC or ODBC connector.
Consider defining a view that represents the flattened version of a table containing complex type columns,
and pointing the JDBC or ODBC application at the view. See Complex Types (CDH 5.5 and higher only) for
details.

Configuring Impala to Work with JDBC


Impala supports the standard JDBC interface, allowing access from commercial Business Intelligence tools and
custom software written in Java or other programming languages. The JDBC driver allows you to access Impala
from a Java program that you write, or a Business Intelligence or similar tool that uses JDBC to communicate
with various database products.
Setting up a JDBC connection to Impala involves the following steps:
Verifying the communication port where the Impala daemons in your cluster are listening for incoming JDBC
requests.
Installing the JDBC driver on every system that runs the JDBC-enabled application.
Specifying a connection string for the JDBC application to access one of the servers running the impalad
daemon, with the appropriate security settings.
Configuring the JDBC Port
The default port used by JDBC 2.0 and later (as well as ODBC 2.x) is 21050. Impala server accepts JDBC connections
through this same port 21050 by default. Make sure this port is available for communication with other hosts
on your network, for example, that it is not blocked by firewall software. If your JDBC client software connects
to a different port, specify that alternative port number with the --hs2_port option when starting impalad.
See Starting Impala for details about Impala startup options. See Ports Used by Impala for information about
all ports used for communication between Impala and clients or between Impala components.
Choosing the JDBC Driver
In Impala 2.0 and later, you have the choice between the Cloudera JDBC Connector and the Hive 0.13 JDBC driver.
Cloudera recommends using the Cloudera JDBC Connector where practical.
If you are already using JDBC applications with an earlier Impala release, you must update your JDBC driver to
one of these choices, because the Hive 0.12 driver that was formerly the only choice is not compatible with
Impala 2.0 and later.
Both the Cloudera JDBC 2.5 Connector and the Hive JDBC driver provide a substantial speed increase for JDBC
applications with Impala 2.0 and higher, for queries that return large result sets.
Complex type considerations:

198 | Cloudera Administration

Managing CDH and Managed Services


The Impala complex types (STRUCT, ARRAY, or MAP) are available in CDH 5.5 / Impala 2.3 and higher. To use these
types with JDBC requires version 2.5.28 or higher of the Cloudera JDBC Connector for Impala. To use these types
with ODBC requires version 2.5.30 or higher of the Cloudera ODBC Connector for Impala. Consider upgrading all
JDBC and ODBC drivers at the same time you upgrade from CDH 5.5 or higher.
Although the result sets from queries involving complex types consist of all scalar values, the queries involve
join notation and column references that might not be understood by a particular JDBC or ODBC connector.
Consider defining a view that represents the flattened version of a table containing complex type columns, and
pointing the JDBC or ODBC application at the view. See Complex Types (CDH 5.5 and higher only) for details.
Enabling Impala JDBC Support on Client Systems
Using the Cloudera JDBC Connector (recommended)
You download and install the Cloudera JDBC 2.5 connector on any Linux, Windows, or Mac system where you
intend to run JDBC-enabled applications. From the Cloudera Connectors download page, you choose the
appropriate protocol (JDBC or ODBC) and target product (Impala or Hive). The ease of downloading and installing
on non-CDH systems makes this connector a convenient choice for organizations with heterogeneous
environments.
Using the Hive JDBC Driver
You install the Hive JDBC driver (hive-jdbc package) through the Linux package manager, on hosts within the
CDH cluster. The driver consists of several Java JAR files. The same driver can be used by Impala and Hive.
To get the JAR files, install the Hive JDBC driver on each CDH-enabled host in the cluster that will run JDBC
applications. Follow the instructions for CDH 5 or CDH 4.
Note: The latest JDBC driver, corresponding to Hive 0.13, provides substantial performance
improvements for Impala queries that return large result sets. Impala 2.0 and later are compatible
with the Hive 0.13 driver. If you already have an older JDBC driver installed, and are running Impala
2.0 or higher, consider upgrading to the latest Hive JDBC driver for best performance with JDBC
applications.
If you are using JDBC-enabled applications on hosts outside the CDH cluster, you cannot use the CDH install
procedure on the non-CDH hosts. Install the JDBC driver on at least one CDH host using the preceding procedure.
Then download the JAR files to each client machine that will use JDBC with Impala:
commons-logging-X.X.X.jar
hadoop-common.jar
hive-common-X.XX.X-cdhX.X.X.jar
hive-jdbc-X.XX.X-cdhX.X.X.jar
hive-metastore-X.XX.X-cdhX.X.X.jar
hive-service-X.XX.X-cdhX.X.X.jar
httpclient-X.X.X.jar
httpcore-X.X.X.jar
libfb303-X.X.X.jar
libthrift-X.X.X.jar
log4j-X.X.XX.jar
slf4j-api-X.X.X.jar
slf4j-logXjXX-X.X.X.jar

To enable JDBC support for Impala on the system where you run the JDBC application:
1. Download the JAR files listed above to each client machine.
Note: For Maven users, see this sample github page for an example of the dependencies you could
add to a pom file instead of downloading the individual JARs.
2. Store the JAR files in a location of your choosing, ideally a directory already referenced in your CLASSPATH
setting. For example:
Cloudera Administration | 199

Managing CDH and Managed Services


On Linux, you might use a location such as /opt/jars/.
On Windows, you might use a subdirectory underneath C:\Program Files.
3. To successfully load the Impala JDBC driver, client programs must be able to locate the associated JAR files.
This often means setting the CLASSPATH for the client process to include the JARs. Consult the documentation
for your JDBC client for more details on how to install new JDBC drivers, but some examples of how to set
CLASSPATH variables include:
On Linux, if you extracted the JARs to /opt/jars/, you might issue the following command to prepend
the JAR files path to an existing classpath:
export CLASSPATH=/opt/jars/*.jar:$CLASSPATH

On Windows, use the System Properties control panel item to modify the Environment Variables for your
system. Modify the environment variables to include the path to which you extracted the files.
Note: If the existing CLASSPATH on your client machine refers to some older version of the
Hive JARs, ensure that the new JARs are the first ones listed. Either put the new JAR files earlier
in the listings, or delete the other references to Hive JAR files.

Establishing JDBC Connections


The JDBC driver class depends on which driver you select.
Note: If your JDBC or ODBC application connects to Impala through a load balancer such as haproxy,
be cautious about reusing the connections. If the load balancer has set up connection timeout values,
either check the connection frequently so that it never sits idle longer than the load balancer timeout
value, or check the connection validity before using it and create a new one if the connection has been
closed.
Using the Cloudera JDBC Connector (recommended)
Depending on the level of the JDBC API your application is targeting, you can use the following fully-qualified
class names (FQCNs):
com.cloudera.impala.jdbc41.Driver
com.cloudera.impala.jdbc41.DataSource
com.cloudera.impala.jdbc4.Driver
com.cloudera.impala.jdbc4.DataSource
com.cloudera.impala.jdbc3.Driver
com.cloudera.impala.jdbc3.DataSource
The connection string has the following format:
jdbc:impala://Host:Port[/Schema];Property1=Value;Property2=Value;...

The port value is typically 21050 for Impala.


For full details about the classes and the connection string (especially the property values available for the
connection string), download the appropriate driver documentation for your platform from the Impala JDBC
Connector download page.
Using the Hive JDBC Driver
For example, with the Hive JDBC driver, the class name is org.apache.hive.jdbc.HiveDriver. Once you have
configured Impala to work with JDBC, you can establish connections between the two. To do so for a cluster that

200 | Cloudera Administration

Managing CDH and Managed Services


does not use Kerberos authentication, use a connection string of the form
jdbc:hive2://host:port/;auth=noSasl. For example, you might use:
jdbc:hive2://myhost.example.com:21050/;auth=noSasl

To connect to an instance of Impala that requires Kerberos authentication, use a connection string of the form
jdbc:hive2://host:port/;principal=principal_name. The principal must be the same user principal you
used when starting Impala. For example, you might use:
jdbc:hive2://myhost.example.com:21050/;principal=impala/myhost.example.com@H2.EXAMPLE.COM

To connect to an instance of Impala that requires LDAP authentication, use a connection string of the form
jdbc:hive2://host:port/db_name;user=ldap_userid;password=ldap_password. For example, you might
use:
jdbc:hive2://myhost.example.com:21050/test_db;user=fred;password=xyz123

Note:
Currently, the Hive JDBC driver does not support connections that use both Kerberos authentication
and SSL encryption. To use both of these security features with Impala through a JDBC application,
use the Cloudera JDBC Connector as the JDBC driver.
Notes about JDBC and ODBC Interaction with Impala SQL Features
Most Impala SQL features work equivalently through the impala-shell interpreter of the JDBC or ODBC APIs.
The following are some exceptions to keep in mind when switching between the interactive shell and applications
using the APIs:
Complex type considerations:
Queries involving the complex types (ARRAY, STRUCT, and MAP) require notation that might not be available
in all levels of JDBC and ODBC drivers. If you have trouble querying such a table due to the driver level or
inability to edit the queries used by the application, you can create a view that exposes a flattened
version of the complex columns and point the application at the view. See Complex Types (CDH 5.5 and
higher only) for details.
The complex types available in CDH 5.5 / Impala 2.3 and higher are supported by the JDBC getColumns()
API. Both MAP and ARRAY are reported as the JDBC SQL Type ARRAY, because this is the closest matching
Java SQL type. This behavior is consistent with Hive. STRUCT types are reported as the JDBC SQL Type
STRUCT.
To be consistent with Hive's behavior, the TYPE_NAME field is populated with the primitive type name
for scalar types, and with the full toSql() for complex types. The resulting type names are somewhat
inconsistent, because nested types are printed differently than top-level types. For example, the following
list shows how toSQL() for Impala types are translated to TYPE_NAME values:
DECIMAL(10,10)
CHAR(10)
VARCHAR(10)
ARRAY<DECIMAL(10,10)>
ARRAY<CHAR(10)>
ARRAY<VARCHAR(10)>

becomes
becomes
becomes
becomes
becomes
becomes

DECIMAL
CHAR
VARCHAR
ARRAY<DECIMAL(10,10)>
ARRAY<CHAR(10)>
ARRAY<VARCHAR(10)>

Managing Key-Value Store Indexer


The Key-Value Store Indexer service uses the Lily HBase Indexer Service to index the stream of records being
added to HBase tables. Indexing allows you to query data stored in HBase with the Solr service.
Cloudera Administration | 201

Managing CDH and Managed Services


The Key-Value Store Indexer service is installed in the same parcel or package along with the CDH 5 or Solr
service.

Adding the Key-Value Store Indexer Service


Required Role:
1.

On the Home > Status tab, click


to the right of the cluster name and select Add a Service. A list of service
types display. You can add one type of service at a time.
2. Select the Key-Value Store Indexer service and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Click Continue.
6. Review the configuration changes to be applied. Confirm the settings entered for file system paths. The file
paths required vary based on the services to be installed. If you chose to add the Sqoop service, indicate
whether to use the default Derby database or the embedded PostgreSQL database. If the latter, type the
database name, host, and user credentials that you specified when you created the database.
Warning: Do not place DataNode data directories on NAS devices. When resizing an NAS, block
replicas can be deleted, which will result in reports of missing blocks.
Click Continue. The wizard starts the services.
7. Click Continue.
8. Click Finish.
Enabling Morphlines with Search and HBase Indexing
Required Role:
Cloudera Morphlines is an open source framework that reduces the time and skills necessary to build or change
Search indexing applications. A morphline is a rich configuration file that simplifies defining an ETL transformation
chain.
1. Go to the Indexer service.
202 | Cloudera Administration

Managing CDH and Managed Services


2.
3.
4.
5.

Click the Configuration tab.


Select Scope > All.
Select Category > Morphlines.
Create the necessary configuration files, and modify the content in the following properties:
Morphlines File Text that goes into the morphlines.conf used by HBase indexers. You should use
$ZK_HOST in this file instead of specifying a ZooKeeper quorum. Cloudera Manager automatically replaces
the $ZK_HOST variable with the correct value during the Solr configuration deployment.
Custom MIME-types File Text that goes verbatim into the custom-mimetypes.xml file used by HBase
Indexers with the detectMimeTypes command. See the Cloudera Morphlines Reference Guide for details
on this command.
Grok Dictionary File Text that goes verbatim into the grok-dictionary.conf file used by HBase
Indexers with the grok command. See the Cloudera Morphlines Reference Guide for details of this
command.
See Extracting, Transforming, and Loading Data With Cloudera Morphlines for information about using
morphlines with Search and HBase.

Managing MapReduce and YARN


CDH supports two versions of the MapReduce computation framework: MRv1 and MRv2, which are implemented
by the MapReduce (MRv1) and YARN (MRv2) services. YARN is backwards-compatible with MapReduce. (All jobs
that run against MapReduce will also run in a YARN cluster).
The MRv2 YARN architecture splits the two primary responsibilities of the JobTracker resource management
and job scheduling/monitoring into separate daemons: a global ResourceManager (RM) and per-application
ApplicationMasters (AM). With MRv2, the ResourceManager (RM) and per-node NodeManagers (NM) form the
data-computation framework. The ResourceManager service effectively replaces the functions of the JobTracker,
and NodeManagers run on worker hosts instead of TaskTracker daemons. The per-application ApplicationMaster
is, in effect, a framework-specific library and negotiates resources from the ResourceManager and works with
the NodeManagers to execute and monitor the tasks. For details of this architecture, see Apache Hadoop NextGen
MapReduce (YARN).
The Cloudera Manager Admin Console has different methods for displaying MapReduce and YARN job history.
See Monitoring MapReduce Jobs and Monitoring YARN Applications.
For information on configuring the MapReduce and YARN services for high availability, see MapReduce (MRv1)
and YARN (MRv2) High Availability on page 315
For information on configuring MapReduce and YARN resource management features, see Resource
Management on page 246.
Defaults and Recommendations
In a Cloudera Manager deployment of a CDH 4 cluster, the MapReduce service is the default MapReduce
computation framework.You can create a YARN service in a CDH 4 cluster, but it is not considered production
ready.
In a Cloudera Manager deployment of a CDH 5 cluster, the YARN service is the default MapReduce computation
framework.In CDH 5, the MapReduce service has been deprecated. However, the MapReduce service is fully
supported for backward compatibility through the CDH 5 lifecycle.
For production uses, Cloudera recommends that only one MapReduce framework should be running at any
given time. If development needs or other use case requires switching between MapReduce and YARN, both
services can be configured at the same time, but only one should be running (to fully optimize the hardware
resources available).
Migrating from MapReduce to YARN
Cloudera Manager provides a wizard described in Importing MapReduce Configurations to YARN on page 209 to
easily migrate MapReduce configurations to YARN. The wizard performs all the steps (Switching Between

Cloudera Administration | 203

Managing CDH and Managed Services


MapReduce and YARN Services on page 204, Updating Dependent Services on page 205, and Configuring Alternatives
Priority on page 205) on this page.
The Activity Monitor role collects information about activities run by the MapReduce service. If MapReduce is
not being used and the reporting data is no longer required, then the Activity Monitor role and database can be
removed:
1. Do one of the following:
Select Clusters > Cloudera Management Service > Cloudera Management Service.
On the Status tab of the Home > Status tab, in Cloudera Management Service table, click the Cloudera
Management Service link.
2.
3.
4.
5.

Click the Instances tab.


Select checkbox for Activity Monitor, select Actions for Selected > Stop, and click Stop to confirm.
Select checkbox for Activity Monitor, select Actions for Selected > Delete, and click Delete to confirm.
Manage the Activity Monitor database. The example below is for a MySQL backend database:
a. Verify the Activity Monitor database:
mysql> show databases;
+--------------------+
| Database
|
+--------------------+
| amon
|
+--------------------+

b. Back up the database:


$ mysqldump -uroot -pcloudera amon > /safe_backup_directory/amon.sql

Drop the database:


mysql> drop database amon;

Once you have migrated to YARN and deleted the MapReduce service, you can remove local data from each
TaskTracker node. The mapred.local.dir parameter is a directory on the local filesystem of each TaskTracker
that contains temporary data for MapReduce. Once the service is stopped, you can remove this directory to free
disk space on each node.
For detailed information on migrating from MapReduce to YARN, see Migrating from MapReduce 1 (MRv1) to
MapReduce 2 (MRv2, YARN).

Switching Between MapReduce and YARN Services


Required Role:
MapReduce and YARN use separate sets of configuration files. No files are removed or altered when you change
to a different framework. To change from YARN to MapReduce (or vice versa):
1.
2.
3.
4.
5.
6.
7.

(Optional) Configure the new MapReduce or YARN service.


Update dependent services to use the chosen framework.
Configure the alternatives priority.
Redeploy the Oozie ShareLib.
Redeploy the client configuration.
Start the framework service to switch to.
(Optional) Stop the unused framework service to free up the resources it uses.

204 | Cloudera Administration

Managing CDH and Managed Services


Updating Dependent Services
Required Role:
When you change the MapReduce framework, the dependent services that must be updated to use the new
framework are:
Hive
Sqoop 2
Oozie
To update a service:
1.
2.
3.
4.
5.
6.
7.

Go to the service.
Click the Configuration tab.
Select Scope > service name(Service Wide).
Select Scope > All.
Locate the MapReduce Service property and select the YARN or MapReduce service.
Click Save Changes to commit the changes.
Select Actions > Restart.

The Hue service is automatically reconfigured to use the same framework as Oozie and Hive. This cannot be
changed. To update the Hue service:
1. Go to the Hue service.
2. Select Actions > Restart.

Configuring Alternatives Priority


Required Role:
The alternatives priority property determines which serviceMapReduce or YARNis used by clients to run
MapReduce jobs. The service with a higher value of the property is used. In CDH 4, the MapReduce service
alternatives priority is set to 92 and the YARN service is set to 91. In CDH 5, the values are reversed; the
MapReduce service alternatives priority is set to 91 and the YARN service is set to 92.
To configure the alternatives priority:
1.
2.
3.
4.
5.
6.
7.
8.

Go to the MapReduce or YARN service.


Click the Configuration tab.
Select Scope > Gateway Default Group.
Select Category > All.
Type Alternatives in Search box.
In the Alternatives Priority property, set the priority value.
Click Save Changes to commit the changes.
Redeploy the client configuration.

Managing MapReduce
For an overview of computation frameworks, insight into their usage and restrictions, and examples of common
tasks they perform, see Managing MapReduce and YARN on page 203.
Configuring the MapReduce Scheduler
Required Role:
The MapReduce service is configured by default to use the FairScheduler. You can change the scheduler type
to FIFO or Capacity Scheduler. You can also modify the Fair Scheduler and Capacity Scheduler configuration. For
further information on schedulers, see Schedulers on page 246.

Cloudera Administration | 205

Managing CDH and Managed Services


Configuring the Task Scheduler Type
1.
2.
3.
4.
5.

Go to the MapReduce service.


Click the Configuration tab.
Select Scope > JobTracker.
Select Category > Classes .
In the Task Scheduler property, select a scheduler.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

6. Click Save Changes to commit the changes.


7. Restart the JobTracker to apply the new configuration:
a. Click the Instances tab.
b. Click the JobTracker role.
c. Select Actions for Selected > Restart.
Modifying the Scheduler Configuration
1.
2.
3.
4.
5.

Go to the MapReduce service.


Click the Configuration tab.
Select Scope > JobTracker.
Select Category > Jobs.
Modify the configuration properties.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

6. Click Save Changes to commit the changes.


7. Restart the JobTracker to apply the new configuration:
a. Click the Instances tab.
b. Click the JobTracker role.
c. Select Actions for Selected > Restart.
Configuring the MapReduce Service to Save Job History
Required Role:
Normally job history is saved on the host on which the JobTracker is running. You can configure JobTracker to
write information about every job that completes to a specified HDFS location. By default, the information is
retained for 7 days.
Enabling Map Reduce Job History To Be Saved to HDFS
1. Create a folder in HDFS to contain the history information. When creating the folder, set the owner and group
to mapred:hadoop with permission setting 775.
2. Go to the MapReduce service.
3. Click the Configuration tab.
4. Select Scope > JobTracker.
5. Select Category > Paths.
6. Set the Completed Job History Location property to the location that you created in step 1.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes.
206 | Cloudera Administration

Managing CDH and Managed Services


8. Restart the MapReduce service.
Setting the Job History Retention Duration
1. Select the JobTracker Default Group category.
2. Set the Job History Files Maximum Age property (mapreduce.jobhistory.max-age-ms) to the length of
time (in milliseconds, seconds, minutes, or hours) that you want job history files to be kept.
3. Restart the MapReduce service.
The Job History Files Cleaner runs at regular intervals to check for job history files that are ready to be deleted.
By default, the interval is 24 hours. To change the frequency with which the Job History Files Cleaner runs:
1. Select the JobTracker Default Group category.
2. Set the Job History Files Cleaner Interval property (mapreduce.jobhistory.cleaner.interval) to the
desired frequency (in milliseconds, seconds, minutes, or hours).
3. Restart the MapReduce service.
Configuring Client Overrides
A configuration property qualified with (Client Override) is a server-side setting that ignores any value a client
tries to set for that property. It performs the same role as its unqualified counterpart, and applies the configuration
to the service with the setting <final>true</final>.
For example, if you set the Map task heap property to 1 GB in the job configuration code, but the service's heap
property qualified with (Client Override) is set to 500 MB, then 500 MB is applied.

Managing YARN
For an overview of computation frameworks, insight into their usage and restrictions, and examples of common
tasks they perform, see Managing MapReduce and YARN on page 203.
Adding the YARN Service
Required Role:
1.

On the Home > Status tab, click


to the right of the cluster name and select Add a Service. A list of service
types display. You can add one type of service at a time.
2. Click the YARN (MR2 Included) radio button and click Continue.
3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Cloudera Administration | 207

Managing CDH and Managed Services


Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.

Configuring Memory Settings for YARN and MRv2


The memory configuration for YARN and MRv2 memory is important to get the best performance from your
cluster. Several different settings are involved. The table below shows the default settings, as well as the settings
that Cloudera recommends, for each configuration option. See Managing MapReduce and YARN on page 203 for
more configuration specifics; and, for detailed tuning advice with sample configurations, see Tuning the Cluster
for MapReduce v2 (YARN).
Table 3: YARN and MRv2 Memory Configuration
Cloudera Manager
Property Name

CDH Property Name

Default Configuration

Cloudera Tuning
Guidelines

Container Memory
Minimum

yarn.scheduler.
1 GB
minimum-allocation-mb

Container Memory
Maximum

yarn.scheduler.
64 GB
maximum-allocation-mb

amount of memory on
largest node

Container Memory
Increment

yarn.scheduler.
512 MB
increment-allocation-mb

Use a fairly large value,


such as 128 MB

Container Memory

yarn.nodemanager.
resource.memory-mb

8 GB

8 GB

Map Task Memory

mapreduce.map.memory.mb 1 GB

1 GB

Reduce Task Memory

mapreduce.reduce.memory.mb 1 GB

1 GB

Map Task Java Opts Base mapreduce.map.java.opts -Djava.net.preferIPv4Stack=true -Djava.net.preferIPv4Stack=true


-Xmx768m
Reduce Task Java Opts
Base

mapreduce.reduce.java.opts -Djava.net.preferIPv4Stack=true -Djava.net.preferIPv4Stack=true

-Xmx768m

ApplicationMaster Memory yarn.app.mapreduce.


am.resource.mb

ApplicationMaster Java
Opts Base

yarn.app.mapreduce.
am.command-opts

1 GB

1 GB

-Djava.net.preferIPv4Stack=true -Djava.net.preferIPv4Stack=true
-Xmx768m

Configuring Directories
Required Role:
Creating the Job History Directory
When adding the YARN service, the Add Service wizard automatically creates a job history directory. If you quit
the Add Service wizard or it does not finish, you can create the directory outside the wizard:
1. Go to the YARN service.
2. Select Actions > Create Job History Dir.
3. Click Create Job History Dir again to confirm.
Creating the NodeManager Remote Application Log Directory
When adding the YARN service, the Add Service wizard automatically creates a remote application log directory.
If you quit the Add Service wizard or it does not finish, you can create the directory outside the wizard:

208 | Cloudera Administration

Managing CDH and Managed Services


1. Go to the YARN service.
2. Select Actions > Create NodeManager Remote Application Log Directory.
3. Click Create NodeManager Remote Application Log Directory again to confirm.
Importing MapReduce Configurations to YARN
Required Role:
Warning: In addition to importing configuration settings, the import process:
Configures services to use YARN as the MapReduce computation framework instead of MapReduce.
Overwrites existing YARN configuration and role assignments.
When you upgrade from CDH 4 to CDH 5, you can import MapReduce configurations to YARN as part of the
upgrade wizard. If you do not import configurations during upgrade, you can manually import the configurations
at a later time:
1. Go to the YARN service page.
2. Stop the YARN service.
3. Select Actions > Import MapReduce Configuration. The import wizard presents a warning letting you know
that it will import your configuration, restart the YARN service and its dependent services, and update the
client configuration.
4. Click Continue to proceed. The next page indicates some additional configuration required by YARN.
5. Verify or modify the configurations and click Continue. The Switch Cluster to MR2 step proceeds.
6. When all steps have been completed, click Finish.
7. (Optional) Remove the MapReduce service.
a. Click the Cloudera Manager logo to return to the Home page.
b.
In the MapReduce row, right-click
and select Delete. Click Delete to confirm.
8. Recompile JARs used in MapReduce applications. For further information, see For MapReduce Programmers:
Writing and Running Jobs.
Configuring the YARN Scheduler
Required Role:
The YARN service is configured by default to use the FairScheduler. You can change the scheduler type to FIFO
or Capacity Scheduler. You can also modify the Fair Scheduler and Capacity Scheduler configuration. For further
information on schedulers, see Schedulers on page 246.
Configuring the Scheduler Type
1.
2.
3.
4.
5.

Go to the YARN service.


Click the Configuration tab.
Select Scope > ResourceManager.
Select Category > Main.
Select a scheduler class.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

6. Click Save Changes to commit the changes.


7. Restart the YARN service.
Modifying the Scheduler Configuration
1. Go to the YARN service.
Cloudera Administration | 209

Managing CDH and Managed Services


2.
3.
4.
5.
6.

Click the Configuration tab.


Click the ResourceManager Default Group category.
Select Scope > ResourceManager.
Type Scheduler in the Search box.
Locate a property and modify the configuration.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

7. Click Save Changes to commit the changes.


8. Restart the YARN service.
Dynamic Resource Management
In addition to the static resource management available to all services, the YARN service also supports dynamic
management of its static allocation. See Dynamic Resource Pools on page 253.
Configuring YARN for Long-running Applications
On a secure cluster, long-running applications such as Spark Streaming jobs will need additional configuration
since the default settings only allow the hdfs user's delegation tokens a maximum lifetime of 7 days, which is
not always sufficient. For instructions on how to work around this issue, see Configuring Spark on YARN for
Long-running Applications.
Task Process Exit Codes
All YARN tasks on the NodeManager are executed in a JVM. When a task executes successfully, the exit code is
0. Exit codes of 0 are not logged, as they are the expected result. Any non-zero exit code is logged as an error.
The non-zero exit code is reported by the NodeManager as an error in the child process. The NodeManager itself
is not affected by the error.
There are multiple reasons why the task JVM might exit with a non-zero code, though there is no exhaustive
list. Exit codes can be split into two categories:
1. Set by the JVM based on the OS signal received by the JVM
2. Directly set in the code
Signal-related exit codes When the OS sends a signal to the JVM, the JVM handles the signal, which could cause
the JVM to exit. Not all signals cause the JVM to exit. Exit codes for OS signals have a value between 128 and
160. Logs show non-zero status codes without further explanation.
Two exit values that typically do not require investigation are 137 and 143. These values are logged when the
JVM is killed by the NodeManager or the OS. The NodeManager might kill a JVM due to task pre-emption (if that
is configured) or speculative execution. The OS might kill the JVM when the JVM exceeds system limits like CPU
time. You should investigate these codes if they appear frequently, as they might indicate a misconfiguration
or a structural problem with regard to resources.
Exit code 154 is used in RecoveredContainerLaunch#call to indicate containers that were lost between
NodeManager restarts without an exit code being recorded. This is usually a bug, and requires investigation.
Other exit codes The JVM might exit if there is an unrecoverable error while executing a task. The exit code and
the message logged should provide more detail. A Java stack trace might also be logged as part of the exit. These
exits should be investigated further to discover a root cause.
In the case of a streaming MapReduce job, the exit code of the JVM is the same as the mapper or reducer in use.
The mapper or reducer can be a shell script or Python script. This means that the underlying script dictates the
exit code: in streaming jobs, you should take this into account during your investigation.

Managing Oozie
This section describes tasks for managing Oozie.

210 | Cloudera Administration

Managing CDH and Managed Services


Configuring Oozie for High Availability
In CDH 5, you can configure multiple active Oozie servers against the same database. Oozie high availability is
active-active or hot-hot so that both Oozie servers are active at the same time, with no failover. High
availability for Oozie is supported in both MRv1 and MRv2 (YARN).
Requirements
The requirements for Oozie high availability are:
Multiple active Oozie servers, preferably identically configured.
JDBC JAR in the same location across all Oozie hosts (for example, /var/lib/oozie/).
External database that supports multiple concurrent connections, preferably with HA support. The default
Derby database does not support multiple concurrent connections.
ZooKeeper ensemble with distributed locks to control database access, and service discovery for log
aggregation.
Load balancer (preferably with HA support, for example HAProxy), virtual IP, or round-robin DNS to provide
a single entry point (of the multiple active servers), and for callbacks from the Application Master or JobTracker.
For information on setting up TLS/SSL communication with Oozie HA enabled, see Additional Considerations
when Configuring TLS/SSL for Oozie HA.
Configuring Oozie High Availability Using Cloudera Manager
Required Role:
Important: Enabling or disabling high availability makes the previous monitoring history unavailable.
Enabling Oozie High Availability
1. Ensure that the requirements are satisfied.
2. In the Cloudera Manager Admin Console, go to the Oozie service.
3. Select Actions > Enable High Availability to see eligible Oozie server hosts. The host running the current
Oozie server is not eligible.
4. Select the host on which to install an additional Oozie server and click Continue.
5. Specify the host and port of the Oozie load balancer, and click Continue. Cloudera Manager stops the Oozie
servers, adds another Oozie server, initializes the Oozie server High Availability state in ZooKeeper, configures
Hue to reference the Oozie load balancer, and restarts the Oozie servers and dependent services.
Disabling Oozie High Availability
1. In the Cloudera Manager Admin Console, go to the Oozie service.
2. Select Actions > Disable High Availability to see all hosts currently running Oozie servers.
3. Select the one host to run the Oozie server and click Continue. Cloudera Manager stops the Oozie service,
removes the additional Oozie servers, configures Hue to reference the Oozie service, and restarts the Oozie
service and dependent services.
Configuring Oozie High Availability Using the Command Line
For installation and configuration instructions for configuring Oozie HA using the command line, see
https://archive.cloudera.com/cdh5/cdh/5/oozie.

Adding the Oozie Service Using Cloudera Manager


Cloudera Manager installs the Oozie service as part of the CDH installation.
You can elect to have the service created and started as part of the Installation wizard. If you elect not to create
the service using the installation wizard, you can use the Add Service wizard to perform the installation. The
wizard will automatically configure and start the dependent services and the Oozie service. See Adding a Service
on page 36 for instructions.
Cloudera Administration | 211

Managing CDH and Managed Services


Redeploying the Oozie ShareLib
Some Oozie actions specifically DistCp, Streaming, Pig, Sqoop, and Hive require external JAR files in order to
run. Instead of having to keep these JAR files in each workflow's lib folder, or forcing you to manually manage
them via the oozie.libpath property on every workflow using one of these actions, Oozie provides the ShareLib.
The ShareLib behaves very similarly to oozie.libpath, except that its specific to the aforementioned actions
and their required JARs.
Redeploying the Oozie ShareLib Using Cloudera Manager
Required Role:
When you upgrade CDH or switch between MapReduce and YARN computation frameworks, redeploy the Oozie
ShareLib as follows:
1.
2.
3.
4.

Go to the Oozie service.


Select Actions > Stop.
Select Actions > Install Oozie ShareLib.
Select Actions > Start.

Redeploying the Oozie ShareLib Using the Command Line


See Installing the Oozie Shared Library in Hadoop HDFS.

Configuring Oozie Data Purge Settings Using Cloudera Manager


All Oozie workflows older than 30 days are purged from the database by default. However, actions associated
with long-running coordinators do not purge until the coordinators complete. If, for example, you schedule a
coordinator to run for a year, all those actions remain in the database for the year.
You can change your Oozie configuration to control when data is purged to improve performance, reduce database
disk usage, or keep the history for a longer period of time. Limiting the size of the Oozie database can also
improve performance during upgrades.
1.
2.
3.
4.

In the Cloudera Manager Admin Console, go to the Oozie service.


Click the Configuration tab.
Type purge in the Search box.
Set the following properties as required for your environment:
Enable Purge for Long-Running Coordinator Jobs
Select this property to enable purging of long-running coordinator jobs for which the workflow jobs are
older than the value you set for the Days to Keep Completed Workflow Jobs property.
Days to Keep Completed Workflow Jobs
Days to Keep Completed Coordinator Jobs
Days to Keep Completed Bundle Jobs

5. Click Save Changes to commit the changes.


6. Select Actions > Restart to restart the Oozie Service.

Adding Schema to Oozie Using Cloudera Manager


Required Role:
For all versions of CDH 4.x, Cloudera Manager configures Oozie with the CDH 4.0.0 schema, even if you are using
a higher version of CDH 4.x. This page explains how to manually add a schema (official or custom) with Cloudera
Manager. Cloudera Manager 5 automatically configures Oozie with all available official schemas per the latest
CDH 5.x release.
1. In the Cloudera Manager Admin Console, go to the Oozie service.
2. Click the Configuration tab.
212 | Cloudera Administration

Managing CDH and Managed Services


3. Select Scope > Oozie Server.
4. Select Category > Advanced.
5. Locate the Oozie SchemaService Workflow Extension Schemas property or search for it by typing its name
in the Search box.
6. Enter the desired schema from Table 4: Oozie Schema - CDH 5 on page 213 or Table 5: Oozie Schema - CDH
4 on page 214, appending .xsd to each entry.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Restart the Oozie service.
Note: Releases are only included in the following tables if a schema was added or removed. If a
release is not in the table, it has the same set of schemas as the previous release that is in the table.
Table 4: Oozie Schema - CDH 5
CDH 5.5.0
distcp

CDH 5.4.0

CDH 5.2.0

CDH 5.1.0

CDH 5.0.1

CDH 5.0.0

distcp-action-0.1 distcp-action-0.1 distcp-action-0.1 distcp-action-0.1 distcp-action-0.1 distcp-action-0.1


distcp-action-0.2 distcp-action-0.2 distcp-action-0.2 distcp-action-0.2 distcp-action-0.2 distcp-action-0.2

email

email-action-0.1 email-action-0.1 email-action-0.1 email-action-0.1 email-action-0.1 email-action-0.1


email-action-0.2 email-action-0.2 email-action-0.2 email-action-0.2

hive

hive-action-0.2 hive-action-0.2 hive-action-0.2 hive-action-0.2 hive-action-0.2 hive-action-0.2


hive-action-0.3 hive-action-0.3 hive-action-0.3 hive-action-0.3 hive-action-0.3 hive-action-0.3
hive-action-0.4 hive-action-0.4 hive-action-0.4 hive-action-0.4 hive-action-0.4 hive-action-0.4
hive-action-0.5 hive-action-0.5 hive-action-0.5 hive-action-0.5 hive-action-0.5 hive-action-0.5
hive-action-0.6

HiveServer2

hive2-action-0.1 hive2-action-0.1 hive2-action-0.1


hive2-action-0.2

oozie-bundle

hive2-action-0.1 hive2-action-0.1 hive2-action-0.1 oozie-bundle-0.1 oozie-bundle-0.1 oozie-bundle-0.1


oozie-bundle-0.2 oozie-bundle-0.2 oozie-bundle-0.2

oozie-coordinator oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1


oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2
oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3
oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4
oozie-sla

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.2

oozie-sla-0.2

oozie-sla-0.2

oozie-sla-0.2

oozie-sla-0.2

oozie-sla-0.2

oozie-workflow oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1 oozie-workflow-0.1


oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2 oozie-workflow-0.2
oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5
oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3 oozie-workflow-0.3

Cloudera Administration | 213

Managing CDH and Managed Services


CDH 5.5.0

CDH 5.4.0

CDH 5.2.0

CDH 5.1.0

CDH 5.0.1

CDH 5.0.0

oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4 oozie-workflow-0.4


oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.4.5 oozie-workflow-0.5
oozie-workflow-0.5 oozie-workflow-0.5 oozie-workflow-0.5 oozie-workflow-0.5 oozie-workflow-0.5
shell

shell-action-0.1 shell-action-0.1 shell-action-0.1 shell-action-0.1 shell-action-0.1 shell-action-0.1


shell-action-0.2 shell-action-0.2 shell-action-0.2 shell-action-0.2 shell-action-0.2 shell-action-0.2
shell-action-0.3 shell-action-0.3 shell-action-0.3 shell-action-0.3 shell-action-0.3 shell-action-0.3

spark

spark-action-0.1 spark-action-0.1

sqoop

sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2 sqoop-action-0.2


sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3 sqoop-action-0.3
sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4 sqoop-action-0.4

ssh

ssh-action-0.1 ssh-action-0.1 ssh-action-0.1 ssh-action-0.1 ssh-action-0.1 ssh-action-0.1


ssh-action-0.2 ssh-action-0.2 ssh-action-0.2 ssh-action-0.2 ssh-action-0.2 ssh-action-0.2

Table 5: Oozie Schema - CDH 4


CDH 4.3.0

CDH 4.2.0

CDH 4.1.0

CDH 4.0.0

distcp-action-0.1

distcp-action-0.1

distcp-action-0.1

distcp-action-0.1

distcp-action-0.2

distcp-action-0.2

email

email-action-0.1

email-action-0.1

email-action-0.1

email-action-0.1

hive

hive-action-0.2

hive-action-0.2

hive-action-0.2

hive-action-0.2

hive-action-0.3

hive-action-0.3

hive-action-0.3

hive-action-0.4

hive-action-0.4

hive-action-0.4

oozie-bundle-0.1

oozie-bundle-0.1

oozie-bundle-0.1

oozie-bundle-0.2

oozie-bundle-0.2

oozie-bundle-0.2

distcp

hive-action-0.5
oozie-bundle

oozie-coordinator

oozie-bundle-0.1

oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1 oozie-coordinator-0.1


oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2 oozie-coordinator-0.2
oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3 oozie-coordinator-0.3
oozie-coordinator-0.4 oozie-coordinator-0.4 oozie-coordinator-0.4

oozie-sla

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.1

oozie-sla-0.1

oozie-workflow

oozie-workflow-0.1

oozie-workflow-0.1

oozie-workflow-0.1

oozie-workflow-0.1

oozie-workflow-0.2

oozie-workflow-0.2

oozie-workflow-0.2

oozie-workflow-0.2

oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5 oozie-workflow-0.2.5

214 | Cloudera Administration

oozie-workflow-0.3

oozie-workflow-0.3

oozie-workflow-0.3

oozie-workflow-0.4

oozie-workflow-0.4

oozie-workflow-0.4

oozie-workflow-0.3

Managing CDH and Managed Services


CDH 4.3.0

CDH 4.2.0

CDH 4.1.0

CDH 4.0.0

shell-action-0.1

shell-action-0.1

shell-action-0.1

shell-action-0.1

shell-action-0.2

shell-action-0.2

shell-action-0.2

shell-action-0.3

shell-action-0.3

shell-action-0.3

sqoop-action-0.2

sqoop-action-0.2

sqoop-action-0.2

sqoop-action-0.3

sqoop-action-0.3

sqoop-action-0.3

sqoop-action-0.4

sqoop-action-0.4

sqoop-action-0.4

ssh-action-0.1

ssh-action-0.1

ssh-action-0.1

oozie-workflow-0.4.5
shell

sqoop

ssh

sqoop-action-0.2

ssh-action-0.1

ssh-action-0.2

Enabling the Oozie Web Console


Enabling the Oozie Web Console Using Cloudera Manager
Required Role:
1.
2.
3.
4.
5.
6.
7.

Download ext-2.2. Extract the contents of the file to /var/lib/oozie/ on the same host as the Oozie Server.
In the Cloudera Manager Admin Console, go to the Oozie service.
Click the Configuration tab.
Locate the Enable Oozie server web console property or search for it by typing its name in the Search box.
Select Enable Oozie server web console.
Click Save Changes to commit the changes.
Restart the Oozie service.

Enabling the Oozie Web Console Using the Command Line


See Enabling the Oozie Web Console.

Setting the Oozie Database Timezone


We recommended that you set the timezone in the Oozie database to GMT. Databases don't handle Daylight
Saving Time (DST) shifts correctly. There might be problems if you run any Coordinators with actions scheduled
to materialize during the one-hour period that gets lost in DST.
To set the timezone in Derby, add the following to CATALINA_OPTS in the oozie-env.sh file:
-Duser.timezone=GMT

To set the timezone just for Oozie in MySQL, add the following argument to
oozie.service.JPAService.jdbc.url:
useLegacyDatetimeCode=false&serverTimezone=GMT

Important: Changing the timezone on an existing Oozie database while Coordinators are already
running might cause Coordinators to shift by the offset of their timezone from GMT one time after
you make this change.
For more information about how to set your database's timezone, see your database's documentation.

Cloudera Administration | 215

Managing CDH and Managed Services


Scheduling in Oozie Using Cron-like Syntax
Most Linux distributions include the cron utility, which is used for scheduling time-based jobs. For example, you
might want cron to run a script that deletes your internet history once a week. This topic details how to schedule
Oozie using Cron-like syntax.
Location
Set the scheduling information in the frequency attribute of the coordinator.xml file. A simple file looks like
the following example. The frequency attribute and scheduling information appear in bold.
<coordinator-app name="MY_APP" frequency="30 14 * *
*" start="2009-01-01T05:00Z" end="2009-01-01T06:00Z" timezone="UTC"
xmlns="uri:oozie:coordinator:0.5">
<action>
<workflow>
<app-path>hdfs://localhost:8020/tmp/workflows</app-path>
</workflow>
</action>
</coordinator-app>

Important: Before CDH 5 Oozie used fixed-frequency scheduling. You could only schedule according
to a set amount of minutes or a set time configured in an EL (Expression Language) function. The
cron-like syntax allows more flexibility.
Syntax and Structure
The cron-like syntax used by Oozie is a string with five space-separated fields:

minute
hour
day-of-month
month
day-of-week

The structure takes the form of * * * * *. For example, 30 14 * * * means that the job runs at at 2:30 p.m.
everyday. The minute field is set to 30, the hour field is set to 14, and the remaining fields are set to *.
Allowed Values and Special Characters
The following table describes special characters allowed and indicates in which fields they can be used.
Table 6: Special Characters
Character

Fields Allowed

Description

* (asterisk)

All

Match all values.

, (comma)

All

Specify multiple values.

- (dash)

All

Specify a range.

/ (forward slash)

All

Specify an increment.

? (question mark)

Day-of-month,
day-of-week

Indicate no specific value (for example, if you want to


specify one but not the other).

Day-of-month,
day-of-week

Indicate the last day of the month or the last day of the
week (Saturday). In the day-of-week field, 6L indicates the
last Friday of the month.

Day-of-month

Indicate the nearest weekday to the given day.

216 | Cloudera Administration

Managing CDH and Managed Services


Character

Fields Allowed

Description

# (pound sign)

Day-of-week

Indicate the nth day of the month

The following table summarizes the valid values for each field.
Field

Allowed Values

Allowed Special Characters

Minute

0-59

, - * /

Hour

0-23

, - * /

Day-of-month

0-31

, - * ? / L W

Month

1-12 or JAN-DEC

, - * /

Day-of-week

1-7 or SUN-SAT

, - * ? / L #

For more information about Oozie cron-like syntax, see Cron syntax in coordinator frequency.
Important: Some cron implementations accept 0-6 as the range for days of the week. Oozie accepts
1-7 instead.
Scheduling Examples
The following examples show cron scheduling in Oozie. Oozies processing time zone is UTC. If you are in a
different time zone, add to or subtract from the appropriate offset in these examples.
Run at the 30th minute of every hour
Set the minute field to 30 and the remaining fields to * so they match every value.
frequency="30 * * * *"

Run at 2:30 p.m. every day


Set the minute field to 30, the hour field to 14, and the remaining fields to *.
frequency="30 14 * * *"

Run at 2:30 p.m. every day in February


Set the minute field to 30, the hour field to 14, the day-of-month field to *, the month field to 2 (February),
and the day-of-week field to *.
frequency="30 14 * 2 *"

Run every 20 minutes between 5:00-10:00 a.m. and between 12:00-2:00 p.m. on the fifth day of each month
Set the minute field to 0/20, the hour field to 5-9,12-14, the day-of-month field to 0/5, and the remaining
fields to *.
frequency="0/20 5-9,12-14 0/5 * *"

Run every Monday at 5:00 a.m.


Set the minute field to 0, the hour field to 5, the day-of-month field to ?, the month field to *, and the
day-of-week field to MON.
frequency="0 5 ? * MON"

Cloudera Administration | 217

Managing CDH and Managed Services


Note: If the ? was set to *, this expression would run the job every day at 5:00 a.m., not just
Mondays.
Run on the last day of every month at 5:00 a.m.
Set the minute field to 0, the hour field to 5, the day-of-month field to L, the month field to *, and the
day-of-week field to ?.
frequency="0 5 L * ?"

Run at 5:00 a.m. on the weekday closest to the 15th day of each month
Set the minute field to 0, the hour field to 5, the day-of-month field to 15W, the month field to *, and the
day-of-week field to ?.
frequency="0 5 15W * ?"

Run every 33 minutes from 9:00-3:00 p.m. on the first Monday of every month
Set the minute field to 0/33, the hour field to 9-14, the day-of-week field to 2#1 (the first Monday), and
the remaining fields to *.
frequency="0/33 9-14 ? * 2#1"

Run every hour from 9:00 a.m.-5:00 p.m. on weekdays


Set the minute field to 0, the hour field to 9-17, the day-of-month field to ?, the month field to *, and
the day-of-week field to 2-6.
frequency="0 9-17 ? * 2-6"

Run on the second-to-last day of every month


Set the minute field to 0, the hour field to 0, the day-of-month field to L-1, the month field to *, and the
day-of-week field to ?.
frequency="0 0 L-1 * ?"

Note: L-1 means the second-to-last day of the month.


Oozie uses Quartz, a job scheduler library, to parse the cron syntax. For more examples, go to the CronTrigger
Tutorial on the Quartz website. Quartz has two fields (second and year) that Oozie does not support.

Managing Solr
You can install the Solr service through the Cloudera Manager installation wizard, using either parcels or packages.
See Installing Search.
You can elect to have the service created and started as part of the Installation wizard. If you elect not to create
the service using the Installation wizard, you can use the Add Service wizard to perform the installation. The
wizard will automatically configure and start the dependent services and the Solr service. See Adding a Service
on page 36 for instructions.
For further information on the Solr service, see:
CDH 4 - Cloudera Search Documentation for CDH 4
CDH 5 - Cloudera Search Guide
The following sections describe how to configure other CDH components to work with the Solr service.

218 | Cloudera Administration

Managing CDH and Managed Services


Configuring the Flume Morphline Solr Sink for Use with the Solr Service
Required Role:
To use a Flume Morphline Solr sink, the Flume service must be running on your cluster. See the Flume Near
Real-Time Indexing Reference (CDH 4) or Flume Near Real-Time Indexing Reference (CDH 5) or information about
the Flume Morphline Solr Sink and Managing Flume on page 77.
1.
2.
3.
4.
5.

Go to the Flume service.


Click the Configuration tab.
Select Scope > Agent
Select Category > Flume-NG Solr Sink.
Edit the following settings, which are templates that you must modify for your deployment:
Morphlines File (morphlines.conf) - Configures Morphlines for Flume agents. You must use $ZK_HOST
in this field instead of specifying a ZooKeeper quorum. Cloudera Manager automatically replaces the
$ZK_HOST variable with the correct value during the Flume configuration deployment.
Custom MIME-types File (custom-mimetypes.xml) - Configuration for the detectMimeTypes command.
See the Cloudera Morphlines Reference Guide for details on this command.
Grok Dictionary File (grok-dictionary.conf) - Configuration for the grok command. See the Cloudera
Morphlines Reference Guide for details on this command.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.

Once configuration is complete, Cloudera Manager automatically deploys the required files to the Flume agent
process directory when it starts the Flume agent. Therefore, you can reference the files in the Flume agent
configuration using their relative path names. For example, you can use the name morphlines.conf to refer
to the location of the Morphlines configuration file.

Deploying Solr with Hue


Required Role:
In CDH 4.3 and earlier, in order to use Solr with Hue, you must update the URL for the Solr Server in the Hue
Server advanced configuration snippet.
1. Go to the Hue service.
2. Click the Configuration tab.
3. Type the word "snippet" in the Search box.
A set of Hue advanced configuration snippet properties displays.
4. Add information about your Solr host to the Hue Server Configuration Advanced Configuration Snippet for
hue_safety_valve_server.ini property. For example, if your hostname is SOLR_HOST, you might add the
following:
[search]
## URL of the Solr Server
solr_url=http://SOLR_HOST:8983/solr

5. Click Save Changes to save your advanced configuration snippet changes.


6. Restart the Hue Service.

Cloudera Administration | 219

Managing CDH and Managed Services


Important: If you are using parcels with CDH 4.3, you must register the "hue-search" application
manually or access will fail. You do not need to do this if you are using CDH 4.4 or later.
1. Stop the Hue service.
2. From the command line do the following:
a.

cd /opt/cloudera/parcels/CDH 4.3.0-1.cdh4.3.0.pXXX/share/hue

(Substitute your own local repository path for the /opt/cloudera/parcels/... if yours is
different, and specify the appropriate name of the CDH 4.3 parcel that exists in your repository.)
b.

./build/env/bin/python ./tools/app_reg/app_reg.py
--install
/opt/cloudera/parcels/SOLR-0.9.0-1.cdh4.3.0.pXXX/share/hue/apps/search

c.

sed -i 's/\.\/apps/..\/..\/..\/..\/..\/apps/g'
./build/env/lib/python2.X/site-packages/hue.pth

where python2.X should be the version you are using (for example, python2.4).
3. Start the Hue service.

Using a Load Balancer with Solr


To configure a load balancer:
1.
2.
3.
4.
5.

Go to the Solr service.


Click the Configuration tab.
Select Scope > Solr
Select Category > All
Enter the hostname and port number of the load balancer in the Solr Load Balancer property in the format
hostname:port number.
Note:
When you set this property, Cloudera Manager regenerates the keytabs for Solr roles. The principal
in these keytabs contains the load balancer hostname.
If there is a Hue service that depends on this Solr service, it also uses the load balancer to
communicate with Solr.

6. Click Save Changes to commit the changes.

Managing Spark
Apache Spark is a general framework for distributed computing that offers very high performance for both
iterative and interactive processing. Apache Spark exposes APIs for Java, Python, R (SparkR), and Scala and
consists of Spark core and several related projects:
Spark SQL - Module for working with structured data. Allows you to seamlessly mix SQL queries with Spark
programs.
Spark Streaming - API that allows you to build scalable fault-tolerant streaming applications.
MLlib - API that implements common machine learning algorithms.
GraphX - API for graphs and graph-parallel computation.
Cloudera supports Spark core, Spark SQL (including DataFrames), Spark Streaming, and MLlib. Cloudera does
not currently offer commercial support for GraphX or SparkR.

220 | Cloudera Administration

Managing CDH and Managed Services


The following Spark SQL features are not supported:
Thrift JDBC/ODBC server
Spark SQL (DataFrames) in PySpark
Spark SQL CLI
The following Spark MLlib features are not supported:
spark.ml
ML pipeline APIs
This section describes how to manage Spark services, use Spark APIs, and configure, build, and run Spark
applications.
To run applications distributed across a cluster, Spark requires a cluster manager. Cloudera supports two cluster
managers: Spark on YARN and Spark Standalone. Cloudera does not support running Spark applications on
Mesos. With Spark on YARN, Spark application processes run on YARN ResourceManager and NodeManager
roles. On Spark Standalone, Spark application processes run on Spark Master and Worker roles.
In CDH 5, Cloudera recommends running Spark applications on a YARN cluster manager instead of on a Spark
Standalone cluster manager. Using YARN as the Spark cluster manager provides several benefits:
You can dynamically share and centrally configure the same pool of cluster resources among all frameworks
that run on YARN.
You can use all the features of YARN schedulers for categorizing, isolating, and prioritizing workloads.
You choose the number of executors to use; in contrast, Spark Standalone requires each application to run
an executor on every host in the cluster.
Spark can run against Kerberos enabled Hadoop clusters and use secure authentication between its processes.
See Spark Authentication.
Related Information

Monitoring Spark Applications


Tuning Spark Applications on page 286
Spark Authentication
Cloudera Spark forum
Apache Spark documentation

Managing Spark Using Cloudera Manager


Spark is available as two services: Spark and Spark (Standalone). In Cloudera Manager 5.1 and lower, the Spark
service runs a Spark Standalone cluster, which has Master and Worker roles. In Cloudera Manager 5.2 and higher,
the service that runs a Spark Standalone cluster has been renamed Spark (Standalone), and the Spark service
runs Spark as a YARN application with only Gateway roles. Both services have a Spark History Server on page
222 role.
You can install, add, and start Spark through the Cloudera Manager Installation wizard using parcels. For more
information, see Installing Spark.
If you do not add the Spark service using the Installation wizard, you can use the Add Service wizard to create
the service. The wizard automatically configures dependent services and the Spark service. For instructions, see
Adding a Service on page 36.
When you upgrade from Cloudera Manager 5.1 or lower to Cloudera 5.2 or higher, Cloudera Manager does not
migrate an existing Spark service, which runs Spark Standalone, to a Spark on YARN service.
For information on Spark applications, see Spark Applications on page 224.

Cloudera Administration | 221

Managing CDH and Managed Services


Managing Spark Standalone Using the Command Line
Important: This item is deprecated and will be removed in a future release. Cloudera supports items
that are deprecated until they are removed. For more information about deprecated and removed
items, see Deprecated Items.
This section describes how to configure and start Spark Standalone services.
For information on installing Spark using the command line, see Spark Installation. For information on configuring
and starting the Spark History Server, see Configuring and Running the Spark History Server Using the Command
Line on page 223.
For information on Spark applications, see Spark Applications on page 224.
Configuring Spark Standalone
Before running Spark Standalone, do the following on every host in the cluster:
Edit /etc/spark/conf/spark-env.sh and change hostname in the last line to the name of the host where
the Spark Master will run:
###
### === IMPORTANT ===
### Change the following to specify the Master host
###
export STANDALONE_SPARK_MASTER_HOST=`hostname`

Optionally, edit other configuration options:


SPARK_MASTER_PORT / SPARK_MASTER_WEBUI_PORT and SPARK_WORKER_PORT /
SPARK_WORKER_WEBUI_PORT, to use non-default ports
SPARK_WORKER_CORES, to set the number of cores to use on this machine
SPARK_WORKER_MEMORY, to set how much memory to use (for example: 1000 MB, 2 GB)
SPARK_WORKER_INSTANCE, to set the number of worker processes per node
SPARK_WORKER_DIR, to set the working directory of worker processes
Starting and Stopping Spark Standalone Clusters
To start Spark Standalone clusters:
1. On one host in the cluster, start the Spark Master:
$ sudo service spark-master start

You can access the Spark Master UI at spark_master:18080.


2. On all the other hosts, start the workers:
$ sudo service spark-worker start

To stop Spark, use the following commands on the appropriate hosts:


$ sudo service spark-worker stop
$ sudo service spark-master stop

Service logs are stored in /var/log/spark.

Spark History Server


The Spark History Server displays information about the history of completed Spark applications. For further
information, see Monitoring Spark Applications.

222 | Cloudera Administration

Managing CDH and Managed Services


For instructions for configuring the Spark History Server to use Kerberos, see Spark Authentication.
Adding the Spark History Server Using Cloudera Manager
By default, the Spark (Standalone) service does not include a History Server. To configure applications to store
history, on Spark clients, set spark.eventLog.enabled to true before starting the application.
Required Role:
To add the History Server:
1.
2.
3.
4.
5.
6.
7.
8.

Go to the Spark service.


Click the Instances tab.
Click the Add Role Instances button.
Select a host in the column under History Server, and then click OK.
Click Continue.
Check the checkbox next to the History Server role.
Select Actions for Selected > Start and click Start.
Click Close when the action completes.

Configuring and Running the Spark History Server Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
1. Create the /user/spark/applicationHistory/ directory in HDFS and set ownership and permissions as
follows:
$
$
$
$

sudo
sudo
sudo
sudo

-u
-u
-u
-u

hdfs
hdfs
hdfs
hdfs

hadoop
hadoop
hadoop
hadoop

fs
fs
fs
fs

-mkdir
-mkdir
-chown
-chmod

/user/spark
/user/spark/applicationHistory
-R spark:spark /user/spark
1777 /user/spark/applicationHistory

2. On hosts from which you will launch Spark jobs, do the following:
a. Create /etc/spark/conf/spark-defaults.conf:
cp /etc/spark/conf/spark-defaults.conf.template
/etc/spark/conf/spark-defaults.conf

b. Add the following to /etc/spark/conf/spark-defaults.conf:


spark.eventLog.dir=hdfs://namenode_host:namenode_port/user/spark/applicationHistory
spark.eventLog.enabled=true

or
spark.eventLog.dir=hdfs://name_service_id/user/spark/applicationHistory
spark.eventLog.enabled=true

c. On one host, start the History Server:


$ sudo service spark-history-server start

Cloudera Administration | 223

Managing CDH and Managed Services


To link the YARN ResourceManager directly to the Spark History Server, set the
spark.yarn.historyServer.address property in /etc/spark/conf/spark-defaults.conf:
spark.yarn.historyServer.address=http://spark_history_server:history_port

By default, history_port is 18088. This causes Spark applications to write their history to the directory that the
History Server reads.

Spark Applications
Spark applications are similar to MapReduce jobs. Each application is a self-contained computation that runs
user-supplied code to compute a result. As with MapReduce jobs, Spark applications can make use of the
resources of multiple hosts.
In MapReduce, the highest-level unit of computation is a job. The system loads the data, applies a map function,
shuffles it, applies a reduce function, and writes it back out to persistent storage. In Spark, the highest-level of
computation is an application, which can run multiple jobs, in sequence or in parallel. A Spark job can consist of
more stages than just a single map and reduce.
To manage the job flow and schedule tasks, Spark relies on a driver process. Typically, this driver process is the
same as the client process used to initiate the job, although in YARN mode, the driver can run on the cluster. In
contrast, in MapReduce, the client process can terminate and the job will continue running.
A Spark application corresponds to an instance of the SparkContext class. An application can be used for a
single batch job, an interactive session with multiple jobs spaced apart, or a long-lived server continually satisfying
requests. Unlike MapReduce, an application will have processes, called executors, running on its behalf even
when it's not running any jobs. This approach enables data storage in memory for quick access, as well as
lightning-fast task startup time.
Spark Terminology
action
Function that returns a value to the driver after running a computation on an RDD.
application
A job, sequence of jobs, or a long-running service issuing new commands as needed or an interactive
exploration session.
ApplicationMaster
A YARN role responsible for negotiating resource requests made by the driver and finding a set of
containers in which to run the Spark application. There is one ApplicationMaster per application.
application JAR
A JAR containing a Spark application. In some cases you can use an "Uber" JAR containing your application
along with its dependencies. The JAR should never include Hadoop or Spark libraries, however, these will
be added at runtime.
cluster manager
An external service for acquiring resources on the cluster: Spark Standalone or YARN.
deploy mode
Identifies where the driver process runs. In client mode, the submitter launches the driver outside of the
cluster. In cluster mode, the framework launches the driver inside the cluster. Client mode is simpler,
but cluster mode allows you to log out after starting a Spark application without terminating the
application.
driver
Process that represents the application session. The driver is responsible for converting the application
to a directed graph of individual steps to execute on the cluster. There is one driver per application.

224 | Cloudera Administration

Managing CDH and Managed Services


executor
A process that serves a Spark application. An executor runs multiple tasks over its lifetime, and multiple
tasks concurrently. A host may have several Spark executors and there are many hosts running Spark
executors for each application.
job
A parallel computation consisting of multiple tasks that gets spawned in response to a Spark action.
partition
A subset of the elements in an RDD. Partitions define the unit of parallelism; Spark processes elements
within a partition in sequence and multiple partitions in parallel. When Spark reads a file from HDFS, it
creates a single partition for a single input split. It returns a single partition for a single block of HDFS
(but the split between partitions is on line split, not the block split), unless you have a compressed text
file. In case of compressed file you would get a single partition for a single file (as compressed text files
are not splittable).
resilient distributed dataset (RDD)
The core programming abstraction in Spark, consisting of a fault-tolerant collection of elements that can
be operated on in parallel.
Spark Standalone
A model of running Spark applications in which a Master daemon coordinates the efforts of Worker
daemons, which run the executors.
Spark on YARN
A model of running Spark applications in which the YARN ResourceManager performs the functions of
the Spark Master. The functions of the Workers are performed by the YARN NodeManagers, which run
the executors.
task
A unit of work on a partition of an RDD. Also referred to as a stage.
transformation
Function that creates a new RDD from an existing RDD.
Configuring Spark Applications
You can specify application configuration parameters as follows:
Pass them directly to the SparkConf that is used to create the SparkContext in your Spark application; for
example:
val conf = new SparkConf().set("spark.dynamicAllocation.initialExecutors", "5")
val sc = new SparkContext(conf)

To avoid hard-coding them in the application:


Pass them using the --conf command-line switch; for example:
spark-submit \
--class com.cloudera.example.YarnExample \
--master yarn \
--deploy-mode cluster \
--conf "spark.eventLog.dir=hdfs:///user/spark/eventlog" \
lib/yarn-example.jar \
10

Specify them in spark-defaults.conf. This file contains lines in the form: "property value". You can
create a comment by putting a hash mark ( # ) at the beginning of a line. You cannot add comments to
the end or middle of a line.

Cloudera Administration | 225

Managing CDH and Managed Services


Here is an example of a spark-defaults.conf file:
spark.master
spark://mysparkmaster.acme.com:7077
spark.eventLog.enabled
true
spark.eventLog.dir
hdfs:///user/spark/eventlog
# Set spark executor memory
spark.executor.memory
2g
spark.logConf
true

It is a good idea to put configuration properties that you want to use for every application into
spark-defaults.conf. See Application Properties for more information.
The order of precedence in configuration parameters is:
1. Parameters passed to SparkConf.
2. Arguments passed to spark-submit, spark-shell, or pyspark.
3. Properties set in spark-defaults.conf.
For more information, see Running Spark on YARN and Spark Configuration.

Configuring Spark Application Properties Using Cloudera Manager


Required Role:
You can configure properties for all Spark applications in spark-defaults.conf as follows:
1.
2.
3.
4.
5.

Go to the Spark service.


Click the Configuration tab.
Select Scope > Gateway.
Select Category > Advanced.
Locate the Spark Client Advanced Configuration Snippet (Safety Valve) for spark-conf/spark-defaults.conf
property.
6. Specify properties described in Application Properties.
If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Deploy the client configuration.

Configuring Spark Application Properties Using the Command Line


To configure properties for all Spark applications using the command line, edit the file
SPARK_HOME/conf/spark-defaults.conf.
Configuring Spark Application Logging Properties
You configure Spark application logging properties in a log4j.properties file.

Configuring Logging Properties Using Cloudera Manager


Required Role:
1.
2.
3.
4.
5.

Go to the Spark service.


Click the Configuration tab.
Select Scope > Gateway.
Select Category > Advanced.
Locate the Spark Client Advanced Configuration Snippet (Safety Valve) for spark-conf/log4j.properties
property.
6. Specify log4j properties.

226 | Cloudera Administration

Managing CDH and Managed Services


If more than one role group applies to this configuration, edit the value for the appropriate role group. See
Modifying Configuration Properties Using Cloudera Manager on page 10.
7. Click Save Changes to commit the changes.
8. Deploy the client configuration.

Configuring Logging Properties Using the Command Line


To specify logging properties using the command line, edit the file SPARK_HOME/conf/log4j.properties.
Building Spark Applications
Best practices when building Spark applications include:
Compiling against the same version of Spark that you are running against, to avoid compatibility issues. For
example, do not assume that applications compiled against Spark 0.9 will run on Spark 1.0 without recompiling.
In addition, some applications compiled under Spark 0.9 or earlier will need changes to their source code to
compile under Spark 1.0. Applications that compile under Spark 1.0 should compile under all future versions.
"Uber" JARs must always be built against the version of Spark you intend to run.
Building a single assembly JAR ("Uber" JAR) that includes all the dependencies. Exclude Spark and Hadoop
classes from the assembly JAR, because they are already on the cluster, and part of the runtime classpath.
In Maven, you can mark the Spark and Hadoop dependencies as provided.
Enabling Fault-Tolerant Processing in Spark Streaming
If the driver host for a Spark Streaming application fails, it can lose data that has been received, but not yet
processed. To ensure that no data is lost, Spark can write out incoming data to HDFS as it is received and use
this data to recover state in the event of a failure. This feature, called Spark Streaming recovery, was introduced
in CDH 5.3 as a Beta feature. Spark Streaming recovery is supported for production use in CDH 5.4.
To enable Spark Streaming recovery:
1. Set the spark.streaming.receiver.writeAheadLog.enable parameter to true in the SparkConf object:
sparkConf.set("spark.streaming.receiver.writeAheadLog.enable", "true")

2. Create a StreamingContext instance using this SparkConf, and specify a checkpoint directory.
3. Use the getOrCreate method in StreamingContext to either create a new context or recover from an old
context from the checkpoint directory:
// Function to create and setup a new StreamingContext
def functionToCreateContext(): StreamingContext = {
val conf = new SparkConf()
sparkConf.set( "spark.streaming.receiver.writeAheadLog.enable", "true ")
val ssc = new StreamingContext(sparkConf,...)
// new context
val kafkaStream = KafkaUtils.createStream(...)
// Do some transformations on the stream....and write it out etc.
ssc.checkpoint(checkpointDirectory)
ssc

// set checkpoint directory

}
// Get StreamingContext from checkpoint data or create a new one
val context = StreamingContext.getOrCreate(checkpointDirectory,
functionToCreateContext _)

For further information, see Checkpointing.


To prevent data loss if a receiver fails, receivers must be able to replay data from the original data sources if
required.
The Kafka receiver automatically replays if the spark.streaming.receiver.writeAheadLog.enable
parameter is set to true.

Cloudera Administration | 227

Managing CDH and Managed Services


The receiver-less Direct Kafka DStream does not require the
spark.streaming.receiver.writeAheadLog.enable parameter, and can function without data loss even
without Streaming recovery.
Both the Flume receivers that come packaged with Spark replay the data automatically on receiver failure.
See Spark Streaming + Kafka Integration Guide and Spark Streaming + Flume Integration Guide.
Using Spark with HBase
A common use case is to use Spark to process data that is destined for HBase. See Importing Data Into HBase
Using Spark.

Using Spark SQL


CDH 5.5.0 adds support for Spark SQL. Spark SQL lets you query structured data inside Spark programs using
either SQL or the DataFrame API.
The following Spark SQL features are not supported:
Thrift JDBC/ODBC server
Spark SQL (DataFrames) in PySpark
Spark SQL CLI
The entry point to all Spark SQL functionality is the SQLContext class or one of its descendants. You create a
SQLContext from a SparkContext. With a SQLContext, you can create a DataFrame from an RDD, a Hive table,
or a data source.
To work with data stored in Hive from Spark applications you construct a HiveContext, which inherits from
SQLContext. Once you have a HiveContext you can access tables in the Hive Metastore and write queries
using HiveQL. When using spark-shell, a HiveContext is already created for you and is available as the
sqlContext variable.
For detailed information on Spark SQL, see the Spark SQL and DataFrame Guide.
Ensuring HiveContext Enforces Secure Access
To ensure that HiveContext enforces ACLs, enable the HDFS-Sentry plug-in as described in Synchronizing
HDFS ACLs and Sentry Permissions. Column-level access control for access from Spark SQL is not supported
by the HDFS-Sentry plug-in.
Using the spark-avro Library to Work with Avro Data Sources
Spark supports loading and saving DataFrames from a variety of data sources. The spark-avro library allows
you to process data encoded in the Avro format using Spark. No Maven pom.xml configuration or other changes
are needed for using spark-avro.
The spark-avro library supports most conversions between Spark SQL and Avro records, making Avro a first-class
citizen in Spark. The library automatically performs all the schema conversion. Spark SQL reads in the data and
converts all of it to Spark's internal representation, so the Avro conversion is only needed during reading and
writing data.
spark-avro supports most conversions between Spark SQL and Avro records. The library automatically does all
the schema conversion. Spark SQL reads in the data and converts all of it to Spark's internal representation, so
that the Avro conversion is only needed during reading and writing data.
Partitioning
spark-avro allows you to easily write and read partitioned data without any extra configuration. Just pass the
columns you want to partition by to the writer as you would for writing Parquet. See Figure 4: Writing Partitioned
Data on page 230 and Figure 5: Reading Partitioned Data on page 231.

228 | Cloudera Administration

Managing CDH and Managed Services


Compression
You can specify the compression to be used on write by modifying the Spark global configuration. The supported
compression types are: uncompressed, snappy, and deflate. You can also specify the level to use with deflate
compression. See Figure 3: Writing Deflate Compressed Records on page 230.
Avro Record Name
You can specify the record name and namespace to use when writing out to disk by passing recordName and
recordNamespace as optional parameters to Spark. See Figure 6: Specifying a Record Name on page 231.
Avro to Spark SQL Conversion
spark-avro supports conversion for all Avro data types, except complex union types:

boolean -> BooleanType


int -> IntegerType
long -> LongType
float -> FloatType
double -> DoubleType
bytes -> BinaryType
string -> StringType
record -> StructType
enum -> StringType
array -> ArrayType
map -> MapType
fixed -> BinaryType

spark-avro supports the following union types:


union(int, long) -> LongType
union(float, double) -> DoubleType
union(any, null) -> any
All doc, aliases, and other fields are stripped when they are loaded into Spark.
Spark SQL to Avro Conversion
Every Spark SQL type is currently supported:

BooleanType -> boolean


IntegerType -> int
LongType -> long
FloatType -> float
DoubleType -> double
BinaryType -> bytes
StringType -> string
StructType -> record
ArrayType -> array
MapType -> map
ByteType -> int
ShortType -> int
DecimalType -> string
BinaryType -> bytes
TimestampType -> long

Cloudera Administration | 229

Managing CDH and Managed Services


Caveats
Since Spark is converting data types, there are a few things to watch out for:
Enumerated types are erased - Avro enumerated types become strings when they are read into Spark. This
is because Spark has no support for enumerated types.
Unions on output - Spark writes everything out as unions of the given type along with a null option.
Avro schema changes - Spark reads everything into an internal representation. This means that even if if all
you do is read and then write the data, the schema for the output will be different.
Spark schema reordering - Spark reorders the elements in its schema when writing them out to disk so that
the elements being partitioned on are the last elements. See Figure 4: Writing Partitioned Data on page 230
for an example.

API Examples
This section provides examples of using the spark-avro API in all the supported languages:
Scala Examples
The easiest way to work with Avro in Spark is to use the DataFrame API. spark-avro has added functionality to
SQLContext to make reading and writing Avro files easier:
import com.databricks.spark.avro._
val sqlContext = new SQLContext(sc)
// The Avro records are converted to Spark types, filtered, and
// then written back out as Avro records
val df = sqlContext.read.avro("input dir")
df.filter("age > 5").write.avro("output dir")

Figure 1: Scala Example with Function


The second way is to provide the class name for the format to use:
import com.databricks.spark.avro._
val sqlContext = new SQLContext(sc)
val df = sqlContext.read.format("com.databricks.spark.avro").load("input dir")
df.filter("age > 5").write.format("com.databricks.spark.avro").save("output dir")

Figure 2: Scala Example with Format


import com.databricks.spark.avro._
val sqlContext = new SQLContext(sc)
// configuration to use deflate compression
sqlContext.setConf("spark.sql.avro.compression.codec", "deflate")
sqlContext.setConf("spark.sql.avro.deflate.level", "5")
val df = sqlContext.read.avro("input dir")
// writes out compressed Avro records
df.write.avro("output dir")

Figure 3: Writing Deflate Compressed Records


import com.databricks.spark.avro._
val sqlContext = new SQLContext(sc)
import sqlContext.implicits._

230 | Cloudera Administration

Managing CDH and Managed Services


val df
(2012,
(2012,
(2012,
(2011,

= Seq(
8, "Batman", 9.8),
8, "Hero", 8.7),
7, "Robot", 5.5),
7, "Git", 2.0)).toDF("year", "month", "title", "rating")

df.write.partitionBy("year", "month").avro("/tmp/output")

Figure 4: Writing Partitioned Data


This code will output a directory structure like this:
-rw-r--r-3 hdfs supergroup
0 2015-11-03 14:58 /tmp/output/_SUCCESS
drwxr-xr-x
- hdfs supergroup
0 2015-11-03 14:58 /tmp/output/year=2011
drwxr-xr-x
- hdfs supergroup
0 2015-11-03 14:58 /tmp/output/year=2011/month=7
-rw-r--r-3 hdfs supergroup
229 2015-11-03 14:58
/tmp/output/year=2011/month=7/part-r-00001-9b89f1bd-7cf8-4ba8-910f-7587c0de5a90.avro
drwxr-xr-x
- hdfs supergroup
0 2015-11-03 14:58 /tmp/output/year=2012
drwxr-xr-x
- hdfs supergroup
0 2015-11-03 14:58 /tmp/output/year=2012/month=7
-rw-r--r-3 hdfs supergroup
231 2015-11-03 14:58
/tmp/output/year=2012/month=7/part-r-00001-9b89f1bd-7cf8-4ba8-910f-7587c0de5a90.avro
drwxr-xr-x
- hdfs supergroup
0 2015-11-03 14:58 /tmp/output/year=2012/month=8
-rw-r--r-3 hdfs supergroup
246 2015-11-03 14:58
/tmp/output/year=2012/month=8/part-r-00000-9b89f1bd-7cf8-4ba8-910f-7587c0de5a90.avro
import com.databricks.spark.avro._
val sqlContext = new SQLContext(sc)
val df = sqlContext.read.avro("/tmp/output")
df.printSchema()
df.filter("year = 2011").foreach(println)

Figure 5: Reading Partitioned Data


This code automatically detects the partitioned data and joins it all up, so it is treated the same as unpartitioned
data. This will also only query the one required directory to decrease disk I/O.
root
|-|-|-|--

title: string (nullable = true)


rating: double (nullable = true)
year: integer (nullable = true)
month: integer (nullable = true)

import com.databricks.spark.avro._
val sqlContext = new SQLContext(sc)
val df = sqlContext.read.avro("input dir")
val name = "AvroTest"
val namespace = "com.cloudera.spark"
val parameters = Map("recordName" -> name, "recordNamespace" -> namespace)
df.write.options(parameters).avro("output dir")

Figure 6: Specifying a Record Name


Java Example
The recommended way to query Avro files in Java is to use the DataFrame API. This code is almost identical to
Figure 2: Scala Example with Format on page 230:
import org.apache.spark.sql.*;
SQLContext sqlContext = new SQLContext(sc);

Cloudera Administration | 231

Managing CDH and Managed Services


// Creates a DataFrame from a file
DataFrame df = sqlContext.read().format("com.databricks.spark.avro").load("input dir");
// Saves the subset of the Avro records read in
df.filter("age > 5").write().format("com.databricks.spark.avro").save("output dir");

Python Example
The recommended way to query Avro files in Python is to use native DataFrame API. The code is almost identical
to Figure 2: Scala Example with Format on page 230:
# Creates a DataFrame from a directory
df = sqlContext.read.format("com.databricks.spark.avro").load("input dir")
# Saves the subset of the Avro records read in
df.where("age > 5").write.format("com.databricks.spark.avro").save("output")

Spark SQL Example


You can also write SQL queries against the DataFrames directly:
CREATE TEMPORARY TABLE table_name
USING com.databricks.spark.avro
OPTIONS (path "input dir")

Using Spark MLlib


CDH 5.5.0 adds support for MLlib, Spark's machine learning library. For information on MLlib, see the Machine
Learning Library (MLlib) Guide.
The following Spark MLlib features are not supported:
spark.ml
ML pipeline APIs
Enabling Native Acceleration For MLlib
MLlib algorithms are compute intensive and benefit from hardware acceleration. To enable native acceleration
for MLlib, perform the following tasks:
Install Required Software
Install the appropriate libgfortran 4.6+ package for your operating system. There is no compatible version
available for RHEL 5 or 6.
OS

Package Name

Package Version

RHEL 7.1

libgfortran

4.8.x

SLES 11 SP3

libgfortran3

4.7.2

Ubuntu 12.04

libgfortran3

4.6.3

Ubuntu 14.04

libgfortran3

4.8.4

Debian 7.1

libgfortran3

4.7.2

Install the GPL Extras parcel or package.


Verify Native Acceleration
You can verify that native acceleration is working only by examining logs after running an application. To verify
with an MLlib example application:

232 | Cloudera Administration

Managing CDH and Managed Services


1. Download MovieLens sample data and copy to HDFS:
$ wget --no-check-certificate \
https://raw.githubusercontent.com/apache/spark/branch-1.5/data/mllib/sample_movielens_data.txt
$ hdfs dfs -copyFromLocal sample_movielens_data.txt /user/hdfs

2. Run the Spark MLlib MovieLens example application:


$ spark-submit --master local --class org.apache.spark.examples.mllib.MovieLensALS
\
$SPARK_HOME/lib/spark-examples.jar \
--rank 5 --numIterations 5 --lambda 1.0 --kryo sample_movielens_data.txt

3. Check the logs. If it hasn't loaded native libraries successfully, you'll see the following four warnings before
the final line, where the RMSE is printed:
15/07/12 12:33:01 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeSystemBLAS
15/07/12 12:33:01 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeRefBLAS
15/07/12 12:33:01 WARN LAPACK: Failed to load implementation from:
com.github.fommil.netlib.NativeSystemLAPACK
15/07/12 12:33:01 WARN LAPACK: Failed to load implementation from:
com.github.fommil.netlib.NativeRefLAPACK
Test RMSE = 1.5378651281107205.

This is what you see on a system with no libgfortran. The same error occurs after installing libgfortran
on RHEL 6 because it installs version 4.4, not 4.6+.
In contrast, on RHEL 7.1, after installing libgfortran 4.8, you should see something like:
15/07/12 13:32:20 WARN BLAS: Failed to load implementation from:
com.github.fommil.netlib.NativeSystemBLAS
15/07/12 13:32:20 WARN LAPACK: Failed to load implementation from:
com.github.fommil.netlib.NativeSystemLAPACK
Test RMSE = 1.5329939324808561.

Using Spark to Read from and Write to the Amazon S3 Filesystem


To read from and write to S3 from Spark, you use Hadoop File APIs (SparkContext.hadoopFile,
JavaHadoopRDD.saveAsHadoopFile, SparkContext.newAPIHadoopRDD,
JavaHadoopRDD.saveAsNewAPIHadoopFile) for input and output of RDDs, providing URLs of the form:
s3a://bucket_name/path/to/file

You can access S3 by the following methods:


Without credentials - Run EC2 instances with instance profiles associated with IAM roles that have the
permissions you want. Any request from a machine with such an profile will authenticate without having to
provide credentials.
With credentials Specify the credentials in a configuration file, such as core-site.xml:
<property>
<name>fs.s3a.access.key>
<value>...</value>
</property>
<property>
<name>fs.s3a.secret.key>
<value>...</value>
</property>

Cloudera Administration | 233

Managing CDH and Managed Services


Specify the credentials at runtime. For example:
sc.hadoopConfiguration.set("fs.s3a.access.key", "...")
sc.hadoopConfiguration.set("fs.s3a.secret.key", "...")

Running Spark Applications


You can run Spark applications locally or distributed across a cluster, either by using an interactive shell or by
submitting an application.
To run applications distributed across a cluster, Spark requires a cluster manager. Cloudera supports two cluster
managers: Spark on YARN and Spark Standalone. Cloudera does not support running Spark applications on
Mesos. With Spark on YARN, Spark application processes run on YARN ResourceManager and NodeManager
roles. On Spark Standalone, Spark application processes run on Spark Master and Worker roles.
In CDH 5, Cloudera recommends running Spark applications on a YARN cluster manager instead of on a Spark
Standalone cluster manager. Using YARN as the Spark cluster manager provides several benefits:
You can dynamically share and centrally configure the same pool of cluster resources among all frameworks
that run on YARN.
You can use all the features of YARN schedulers for categorizing, isolating, and prioritizing workloads.
You choose the number of executors to use; in contrast, Spark Standalone requires each application to run
an executor on every host in the cluster.
Spark can run against Kerberos enabled Hadoop clusters and use secure authentication between its processes.
See Spark Authentication.
For information on monitoring Spark applications, see Monitoring Spark Applications.
Running Spark Applications Interactively
You can run a Spark application interactively using the the Scala spark-shell or Python pyspark shell application.
For a complete list of options, run spark-shell or Python pyspark with the -h flag.
For example, when the shell is running, you can perform a word-count application using the following code
examples:
Scala
val file = sc.textFile("hdfs://namenode:8020/path/to/input")
val counts = file.flatMap(line => line.split(" ")).map(word => (word,
1)).reduceByKey(_ + _)
counts.saveAsTextFile("hdfs://namenode:8020/user/hdfs/output")

Python
file = sc.textFile("hdfs://namenode:8020/path/to/input")
counts = file.flatMap(lambda line: line.split(" ")).map(lambda word: (word,
1)).reduceByKey(lambda v1,v2: v1 + v2)
counts.saveAsTextFile("hdfs://namenode:8020/user/hdfs/output")

Note: Some applications run in an interactive shell that have nested definitions may encounter a
Task not serializable exception, because of a limitation in the way Scala compiles code. Cloudera
recommends submitting these applications to a cluster.

234 | Cloudera Administration

Managing CDH and Managed Services


Submitting Spark Applications
Using the spark-submit Script
You can submit compiled Spark applications with the spark-submit script.
spark-submit \
--option value
application jar | python file\
[application arguments]

Example: Running SparkPi on YARN on page 239 and Example: Running SparkPi on Spark Standalone on page
242 demonstrate how to run a sample application, SparkPi, which is packaged with Spark. It computes an
approximation to the value of Pi.
Table 7: spark-submit Options
Option

Description

application jar

Path to a bundled JAR file including your application and all dependencies.
The URL must be globally visible inside your cluster; for example, an
hdfs:// path or a file:// path that exists on all nodes.

python file

Path to a Python file containing your application. The URL must be globally
visible inside your cluster; for example, an hdfs:// path or a file://
path that exists on all nodes.

--py-files python files

Comma-separated list of .zip, .egg, or .py files to place on the PYTHONPATH


for Python apps.

--files files

Comma-separated list of files to be placed in the working directory of


each executor.

application arguments

Arguments passed to the main method of your main class, if any.

Command-Line Options
You specify command-line options using the form --option value instead of --option=value. (Use a space
instead of an equals sign.)
Table 8: Options
Option

Description

--class

The FQCN of the class containing the main method of the application. For
example, org.apache.spark.examples.SparkPi.

--conf

Spark configuration property in key=value format. For values that contain


spaces, surround "key=value" with quotes (as shown).

--deploy-mode

The deployment mode: cluster and client. In cluster mode the driver runs
on worker hosts. In client mode, the driver runs locally as an external
client. Broadly, cluster mode should be used production jobs, while client
mode is more appropriate for interactive and debugging uses, where you
want to see your application's output immediately. For affect of the
deployment mode when running on YARN, see Deployment Modes on
page 237.
Default: client.

--driver-cores

Number of cores used by the driver, only in cluster mode.


Default: 1.
Cloudera Administration | 235

Managing CDH and Managed Services


Option

Description

--driver-memory

The maximum heap size (represented as a JVM string; for example 1024m,
2g, and so on) to allocate to the driver. Alternatively, you can use the
spark.driver.memory configuration parameter.

--jars

Additional JARs to be loaded into the classpath of drivers and executors


in cluster mode or into the executor classpath in client mode. These JARs
can be in the HDFS file system; otherwise they must be available locally
on each executor. The path to the JARs on HDFS must be specified as
hdfs://nameservice:8020/path/to/jar.

--master

The location to run the application.

--packages

Comma-separated list of Maven coordinates of JARs to include on the


driver and executor classpaths. The local Maven repo, then Maven central,
and any additional remote repositories specified in --repositories are
searched in that order. The format for the coordinates should be
groupId:artifactId:version.

--repositories

Comma-separated list of additional remote repositories to search for the


Maven coordinates given with --packages.

Table 9: Master Values


Master

Description

local

Run Spark locally with one worker thread (that is, no parallelism).

local[K]

Run Spark locally with K worker threads (ideally, set this to the number
of cores on your host).

local[*]

Run Spark locally with as many worker threads as logical cores on your
host.

spark://host:port

Run on the Spark Standalone Master on the specified host. The port must
be the one your Master is configured to use (7077 by default).

yarn

Run on a YARN cluster. The cluster location is determined by the


HADOOP_CONF_DIR or YARN_CONF_DIR variable. See Configuring the
Environment on page 239.

Cluster Overview
MapReduce runs each task in its own process. When a task completes, the process terminates. In Spark, many
tasks can run concurrently in an executor, and the executor exists for the lifetime of the Spark application, even
when no jobs are running. A cluster manager starts the executor processes.
In this model, tasks can start very quickly and process in-memory data. However, you have less control of
resource management. Because the number of executors for an application is typically fixed, and each executor
has a fixed allotment of resources, an application uses the same amount of resources for the full duration that
it runs.
When running on YARN, you can dynamically increase and decrease the number of executors.
Following are the steps that occur when you submit a Spark application to a cluster:
1.
2.
3.
4.

spark-submit launches the driver program and invokes the main method in the Spark application.

The driver program requests resources from the cluster manager to launch executors.
The cluster manager launches executors on behalf of the driver program.
The driver process runs the user application. Based on the resilient distributed dataset (RDD) actions and
transformations in the program, the driver sends tasks to executors.

236 | Cloudera Administration

Managing CDH and Managed Services


5. Tasks are run on executor processes to compute and save results.
6. If the driver's main method exits or calls SparkContext.stop, it terminates the executors and releases
resources from the cluster manager.
Table 10: Spark Runtime Summary
Mode

YARN Client Mode

YARN Cluster Mode

Spark Standalone

Driver runs in

Client

ApplicationMaster

Client

Requests resources

ApplicationMaster

ApplicationMaster

Client

Starts executor processes YARN NodeManager

YARN NodeManager

Spark Worker

Persistent services

YARN ResourceManager
and NodeManagers

YARN ResourceManager
and NodeManagers

Spark Master and Workers

Supports Spark Shell

Yes

No

Yes

For more information, see Cluster Mode Overview.


Running Spark Applications on YARN
When Spark applications run on YARN, resource management, scheduling, and security are controlled by YARN.
You can run an application in cluster mode or client mode.
When running Spark on YARN, each Spark executor runs as a YARN container. Where MapReduce schedules a
container and starts a JVM for each task, Spark hosts multiple tasks within the same container. This approach
enables several orders of magnitude faster task startup time.

Deployment Modes
In YARN, each application instance has an ApplicationMaster process, which is the first container started for
that application. The application is responsible for requesting resources from the ResourceManager, and, when
allocated them, instructing NodeManagers to start containers on its behalf. ApplicationMasters obviate the
need for an active client the process starting the application can terminate and coordination continues from
a process managed by YARN running on the cluster.
For the option to specify the deployment mode, see Command-Line Options on page 235.
Cluster Deployment Mode
In cluster mode, the driver runs in the ApplicationMaster on a cluster host chosen by YARN. This means that
the same process, which runs in a YARN container, is responsible for both driving the application and requesting
resources from YARN. The client that launches the application doesn't need to continue running for the entire
lifetime of the application.

Cloudera Administration | 237

Managing CDH and Managed Services

Cluster mode is not well suited to using Spark interactively. Spark applications that require user input, such as
spark-shell and pyspark, need the Spark driver to run inside the client process that initiates the Spark
application.
Client Deployment Mode
In client mode, the driver runs on the host where the job is submitted. The ApplicationMaster is merely present
to request executor containers from YARN. The client communicates with those containers to schedule work
after they start:

238 | Cloudera Administration

Managing CDH and Managed Services


Configuring the Environment
Spark requires the HADOOP_CONF_DIR or YARN_CONF_DIR environment variable point to the directory containing
the client-side configuration files for the cluster. These configurations are used to write to HDFS and connect
to the YARN ResourceManager. If you are using a Cloudera Manager deployment, these variables are configured
automatically. If you are using an unmanaged deployment, ensure that you set the variables as described in
Running Spark on YARN.

Running a Spark Shell Application on YARN


To run the spark-shell or pyspark client on YARN, use the --master yarn --deploy-mode client flags
when you start the application.
If you are using a Cloudera Manager deployment, these properties are configured automatically.

Example: Running SparkPi on YARN


Running SparkPi in YARN Cluster Mode
To run SparkPi in cluster mode:
CDH 5.2 and lower
spark-submit --class org.apache.spark.examples.SparkPi --master yarn \
--deploy-mode cluster SPARK_HOME/examples/lib/spark-examples.jar 10

CDH 5.3 and higher


spark-submit --class org.apache.spark.examples.SparkPi --master yarn\
--deploy-mode cluster SPARK_HOME/lib/spark-examples.jar 10

The argument passed after the JAR controls how close to Pi the approximation should be.
The command will continue to print out status until the job finishes or you press control-C. Terminating the
spark-submit process in cluster mode does not terminate the Spark application as it does in client mode. To
monitor the status of the running application, run yarn application -list.
Running SparkPi in YARN Client Mode
To run SparkPi in client mode:
CDH 5.2 and lower
spark-submit --class org.apache.spark.examples.SparkPi --master yarn \
--deploy-mode client SPARK_HOME/examples/lib/spark-examples.jar 10

CDH 5.3 and higher


spark-submit --class org.apache.spark.examples.SparkPi --master yarn \
--deploy-mode client SPARK_HOME/lib/spark-examples.jar 10

The argument passed after the JAR controls how close to Pi the approximation should be.

Configuring Spark on YARN Applications


As well as the class implementing the application and the library containing the class, the options applicable to
running Spark applications on YARN are:
Option

Description

--archives

Comma separated list of archives to be extracted into the working directory


of each executor.

--driver-memory

The maximum size of each Spark driver's Java heap memory. Alternatively,
you can use the spark.driver.memory configuration property.
Cloudera Administration | 239

Managing CDH and Managed Services


Option

Description

--executor-cores

Number of processor cores to allocate on each executor. Alternatively,


you can use the spark.executor.cores configuration property.

--executor-memory

The maximum heap size to allocate to each executor. Alternatively, you


can use the spark.executor.memory configuration property.

--num-executors

The total number of YARN containers to allocate for this application.


Alternatively, you can use the spark.executor.instances configuration
property.

--queue

The YARN queue to submit to. For further information, see Assigning
Applications and Queries to Resource Pools on page 257.
Default: default.

During initial installation, Cloudera Manager tunes properties according to your cluster environment.
In addition to the command-line options, the following properties are available:
Property

Description

spark.yarn.executor.memoryOverhead The amount of extra off-heap memory that can be requested from YARN,
per executor process. Combined with spark.executor.memory, this is

the total memory YARN can use to create a JVM for an executor process.
spark.yarn.driver.memoryOverhead The amount of extra off-heap memory that can be requested from YARN
per driver. Combined with spark.driver.memory, this is the total memory

that YARN can use to create a JVM for a driver process.

Dynamic Allocation
Spark can dynamically scale the number of executors allocated to the application up and down based on the
workload. Dynamic allocation is useful when multiple applications share resources in a Spark cluster. When
dynamic allocation is enabled, Spark adds and removes executors dynamically to and from jobs. When an
application becomes idle, its resources can be released to the resource pool and acquired by other applications.
Dynamic allocation is enabled by setting spark.executor.dynamicAllocation.enabled to true. This is the
default value. The following table describes additional dynamic allocation properties:
Property

Description

spark.executor.dynamicAllocation.initialExecutors The initial number of executors for a Spark application when dynamic

allocation is enabled.
Default: 1.
spark.executor.dynamicAllocation.minExecutors The lower bound for the number of executors.

Default: 0.
spark.executor.dynamicAllocation.maxExecutors The upper bound for the number of executors.

Default: Integer.MAX_VALUE.
If you are using Spark Streaming, Cloudera recommends that you disable dynamic allocation by setting
spark.dynamicAllocation.enabled to false when running streaming applications.
If you use the --num-executors command line argument or the spark.executor.instances property, dynamic
allocation is disabled. For more information, see dynamic resource allocation.

240 | Cloudera Administration

Managing CDH and Managed Services


Configuring Authentication for Long-Running Spark Streaming Jobs
If you are using authenticated Spark communication you must perform additional configuration steps when
running long-running Spark Streaming jobs. See Configuring Spark on YARN for Long-running Applications.

Optimizing YARN Mode


Unmanaged CDH Deployments
In CDH deployments not managed by Cloudera Manager, Spark copies the Spark assembly JAR file to HDFS each
time you run spark-submit. You can avoid doing this copy in one of the following ways:
Set spark.yarn.jar to the local path to the assembly JAR:
package installation - local:/usr/lib/spark/lib/spark-assembly.jar
parcel installation - local:/opt/cloudera/parcels/CDH/lib/spark/lib/spark-assembly.jar
This is the default for Cloudera Manager deployments.
Upload the JAR and configure the JAR location:
1. Manually upload the Spark assembly JAR file to HDFS:
$ hdfs dfs -mkdir -p /user/spark/share/lib
$ hdfs dfs -put SPARK_HOME/assembly/lib/spark-assembly_*.jar
/user/spark/share/lib/spark-assembly.jar

You must manually upload the JAR each time you upgrade Spark to a new minor CDH release.
2. Set spark.yarn.jar to the HDFS path:
spark.yarn.jar=hdfs://namenode:8020/user/spark/share/lib/spark-assembly.jar

Running Spark Applications on Spark Standalone


Important: This item is deprecated and will be removed in a future release. Cloudera supports items
that are deprecated until they are removed. For more information about deprecated and removed
items, see Deprecated Items.
In CDH 5, Cloudera recommends running Spark applications on a YARN cluster manager instead of on a Spark
Standalone cluster manager. Using YARN as the Spark cluster manager provides several benefits:
You can dynamically share and centrally configure the same pool of cluster resources among all frameworks
that run on YARN.
You can use all the features of YARN schedulers for categorizing, isolating, and prioritizing workloads.
You choose the number of executors to use; in contrast, Spark Standalone requires each application to run
an executor on every host in the cluster.
Spark can run against Kerberos enabled Hadoop clusters and use secure authentication between its processes.
See Spark Authentication.

Running a Spark Shell Application on Spark Standalone


To run the spark-shell or pyspark application on Spark Standalone, use the --master
http://spark_master:spark_master_port flag when you start the application.

Submitting Spark Applications to Spark Standalone


To submit a Spark application on Spark Standalone, supply the --master and --deploy-mode client arguments
to spark-submit.

Cloudera Administration | 241

Managing CDH and Managed Services


Example: Running SparkPi on Spark Standalone
CDH 5.2 and lower
spark-submit --class org.apache.spark.examples.SparkPi --deploy-mode client \
--master http://spark_master:spark_master_port \
SPARK_HOME/examples/lib/spark-examples.jar 10

CDH 5.3 and higher


spark-submit --class org.apache.spark.examples.SparkPi --deploy-mode client \
--master http://spark_master:spark_master_port \
SPARK_HOME/lib/spark-examples.jar 10

The argument passed after the JAR controls how close to Pi the approximation should be.

Running a Crunch Application with Spark


The blog post How-to: Run a Simple Apache Spark App in CDH 5 provides a tutorial on writing, compiling, and
running a Spark application. Taking that article as a starting point, do the following to run a Crunch application
with Spark.
1. Add both the crunch-core and crunch-spark dependencies to your Maven project, along with the other
dependencies shown in the blog post.
2. Use the SparkPipeline (org.apache.crunch.impl.spark.SparkPipeline) where you would have used
the MRPipeline instance in the declaration of your Crunch pipeline. The SparkPipeline will need either a
String that contains the connection string for the Spark master (local for local mode, yarn-client for
YARN) or an actual JavaSparkContext instance.
3. Update the SPARK_SUBMIT_CLASSPATH:
export
SPARK_SUBMIT_CLASSPATH=./commons-codec-1.4.jar:$SPARK_HOME/assembly/lib/*:./myapp-jar-with-dependencies.jar

Important: The commons-codec-1.4 dependency must come before the SPARK_HOME dependencies.
4. Now you can start the pipeline using your Crunch application jar-with-dependencies file using the
spark-submit script, just as you would for a regular Spark pipeline.

Managing the Sqoop 1 Client


The Sqoop 1 client allows you to create a Sqoop 1 gateway and deploy the client configuration.
Installing JDBC Drivers
Sqoop 1 does not ship with third-party JDBC drivers; you must download them separately. For information on
downloading and saving the drivers, see (CDH 4) Installing JDBC Drivers and (CDH 5) Installing JDBC Drivers.
Ensure that you do not save JARs in the CDH parcel directory /opt/cloudera/parcels/CDH, because this
directory is overwritten when you upgrade CDH.
Adding the Sqoop 1 Client
Required Role:
The Sqoop 1 client packages are installed by the Installation wizard. However, the client configuration is not
deployed. To create a Sqoop 1 gateway and deploy the client configuration:
1.

On the Home > Status tab, click


to the right of the cluster name and select Add a Service. A list of service
types display. You can add one type of service at a time.
2. Select the Sqoop 1 Client service and click Continue.
242 | Cloudera Administration

Managing CDH and Managed Services


3. Select the radio button next to the services on which the new service should depend. All services must depend
on the same ZooKeeper service. Click Continue.
4. Customize the assignment of role instances to hosts. The wizard evaluates the hardware configurations of
the hosts to determine the best hosts for each role. The wizard assigns all worker roles to the same set of
hosts to which the HDFS DataNode role is assigned. You can reassign role instances if necessary.
Click a field below a role to display a dialog containing a list of hosts. If you click a field containing multiple
hosts, you can also select All Hosts to assign the role to all hosts, or Custom to display the pageable hosts
dialog.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
Click the View By Host button for an overview of the role assignment by hostname ranges.
5. Click Continue. The client configuration deployment command runs.
6. Click Continue and click Finish.

Managing Sqoop 2
Cloudera Manager can install the Sqoop 2 service as part of the CDH installation.
Note: Moving from Sqoop 1 to Sqoop 2: Sqoop 2 is essentially the future of the Apache Sqoop project.
However, since Sqoop 2 currently lacks some of the features of Sqoop 1, Cloudera recommends you
use Sqoop 2 only if it contains all the features required for your use case; otherwise, continue to use
Sqoop 1.
You can elect to have the service created and started as part of the Installation wizard if you choose to add it
in Custom Services. If you elect not to create the service using the Installation wizard, you can use the Add
Service wizard to perform the installation. The wizard will automatically configure and start the dependent
services and the Sqoop 2 service. See Adding a Service on page 36 for instructions.
Installing JDBC Drivers
The Sqoop 2 service does not ship with third-party JDBC drivers; you must download them separately. For
information on downloading and saving the drivers, see (CDH 4) Configuring Sqoop 2 and (CDH 5) Configuring
Sqoop 2. Ensure that you do not save JARs in the CDH parcel directory /opt/cloudera/parcels/CDH, because
this directory is overwritten when you upgrade CDH.

Managing ZooKeeper
Required Role:
When adding the ZooKeeper service, the Add Service wizard automatically initializes the data directories. If you
quit the Add Service wizard or it does not finish successfully, you can initialize the directories outside the wizard
by doing these steps:
1. Go to the ZooKeeper service.
Cloudera Administration | 243

Managing CDH and Managed Services


2. Select Actions > Initialize.
3. Click Initialize again to confirm.
Note: If the data directories are not initialized, the ZooKeeper servers cannot be started.
In a production environment, you should deploy ZooKeeper as an ensemble with an odd number of servers. As
long as a majority of the servers in the ensemble are available, the ZooKeeper service will be available. The
minimum recommended ensemble size is three ZooKeeper servers, and Cloudera recommends that each server
run on a separate machine. In addition, the ZooKeeper server process should have its own dedicated disk storage
if possible.

Configuring Services to Use the GPL Extras Parcel


After you install the GPL Extras parcel, reconfigure and restart services that need to use LZO functionality. Any
service that does not require the use of LZO need not be configured.
HDFS
1.
2.
3.
4.
5.

Go to the HDFS service.


Click the Configuration tab.
Search for the io.compression.codecs property.
In the Compression Codecs property, click in the field, then click the + sign to open a new value field.
Add the following two codecs:
com.hadoop.compression.lzo.LzoCodec
com.hadoop.compression.lzo.LzopCodec

6. Save your configuration changes.


7. Restart HDFS.
8. Redeploy the HDFS client configuration.
Oozie
1. Go to /var/lib/oozie on each Oozie server and even if the LZO JAR is present, symlink the Hadoop LZO
JAR:
CDH 5 - /opt/cloudera/parcels/GPLEXTRAS/lib/hadoop/lib/hadoop-lzo.jar
CDH 4 - /opt/cloudera/parcels/HADOOP_LZO/lib/hadoop/lib/hadoop-lzo.jar
2. Restart Oozie.
HBase
Restart HBase.
Impala
Restart Impala.
Hive
Restart the Hive server.
Sqoop 2
1. Add the following entries to the Sqoop Service Environment Advanced Configuration Snippet:
HADOOP_CLASSPATH=$HADOOP_CLASSPATH:/opt/cloudera/parcels/GPLEXTRAS/lib/hadoop/lib/*
JAVA_LIBRARY_PATH=$JAVA_LIBRARY_PATH:/opt/cloudera/parcels/GPLEXTRAS/lib/hadoop/lib/native

244 | Cloudera Administration

Managing CDH and Managed Services


2. Restart the Sqoop service.

Cloudera Administration | 245

Resource Management

Resource Management
Resource management helps ensure predictable behavior by defining the impact of different services on cluster
resources. The goals of resource management features are to:
Guarantee completion in a reasonable time frame for critical workloads
Support reasonable cluster scheduling between groups of users based on fair allocation of resources per
group
Prevent users from depriving other users access to the cluster

Schedulers
A scheduler is responsible for deciding which tasks get to run and where and when to run them. The MapReduce
and YARN computation frameworks support the following schedulers:
FIFO - Allocates resources based on arrival time.
Fair - Allocates resources to weighted pools, with fair sharing within each pool.
CDH 4 Fair Scheduler
CDH 5 Fair Scheduler
Capacity - Allocates resources to pools, with FIFO scheduling within each pool.
CDH 4 Capacity Scheduler
CDH 5 Capacity Scheduler
The scheduler defaults for MapReduce and YARN are:
MapReduce - Cloudera Manager, CDH 5, and CDH 4 set the default scheduler to FIFO. FIFO is set as the default
for backward-compatibility purposes, but Cloudera recommends Fair Scheduler because Impala and Llama
are optimized for it. Capacity Scheduler is also available.
If you are running CDH 4, you can specify how MapReduce jobs share resources by configuring the MapReduce
scheduler.
YARN - Cloudera Manager, CDH 5, and CDH 4 set the default to Fair Scheduler. Cloudera recommends Fair
Scheduler because Impala and Llama are optimized for it. FIFO and Capacity Scheduler are also available.
In YARN, the scheduler is responsible for allocating resources to the various running applications subject to
familiar constraints of capacities, queues, and so on. The scheduler performs its scheduling function based
the resource requirements of the applications; it does so based on the abstract notion of a resource container
that incorporates elements such as memory, CPU, disk, network, and so on.
The YARN scheduler has a pluggable policy plug-in, which is responsible for partitioning the cluster resources
among the various queues, applications, and so on.
If you are running CDH 5, you can specify how YARN applications share resources by manually configuring
the YARN scheduler. Alternatively you can use Cloudera Manager dynamic allocation features to manage
the scheduler configuration.

Cloudera Manager Resource Management Features


Cloudera Manager provides the following features to assist you with allocating cluster resources to services.

246 | Cloudera Administration

Resource Management
Static Allocation
Cloudera Manager 4 introduced the ability to partition resources across HBase, HDFS, Impala, MapReduce, and
YARN services by allowing you to set configuration properties that were enforced by Linux control groups (Linux
cgroups). With Cloudera Manager 5, the ability to statically allocate resources using cgroups is configurable
through a single static service pool wizard. You allocate services a percentage of total resources and the wizard
configures the cgroups.
For example, the following figure illustrates static pools for HBase, HDFS, Impala, and YARN services that are
respectively assigned 20%, 30%, 20%, and 30% of cluster resources.

Dynamic Allocation
Cloudera Manager allows you to manage mechanisms for dynamically apportioning resources statically allocated
to YARN and Impala using dynamic resource pools.
Depending on the version of CDH you are using, dynamic resource pools in Cloudera Manager support the
following resource management (RM) scenarios:
(CDH 5) YARN Independent RM - YARN manages the virtual cores, memory, running applications, and scheduling
policy for each pool. In the preceding diagram, three dynamic resource pools - Dev, Product, and Mktg with
weights 3, 2, and 1 respectively - are defined for YARN. If an application starts and is assigned to the Product
pool, and other applications are using the Dev and Mktg pools, the Product resource pool will receive 30% x
2/6 (or 10%) of the total cluster resources. If there are no applications using the Dev and Mktg pools, the
YARN Product pool will be allocated 30% of the cluster resources.
(CDH 5) YARN and Impala Independent RM - YARN manages the virtual cores, memory, running applications,
and scheduling policy for each pool; Impala manages memory for pools running queries and limits the number
of running and queued queries in each pool.
(CDH 5 and CDH 4) Impala Independent RM - Impala manages memory for pools running queries and limits
the number of running and queued queries in each pool.
(CDH 5) YARN and Impala Integrated RM Note: Though Impala can be used together with YARN via simple configuration of Static Service
Pools in Cloudera Manager, the use of the general-purpose component Llama for integrated
resource management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
YARN manages memory for pools running Impala queries; Impala limits the number of running and queued
queries in each pool. In the YARN and Impala integrated RM scenario, Impala services can reserve resources
through YARN, effectively sharing the static YARN service pool and resource pools with YARN applications.
The integrated resource management scenario, where both YARN and Impala use the YARN resource
management framework, require the Impala Llama role.
In the following figure, the YARN and Impala services have a 50% static share which is subdivided among the
original resource pools with an additional resource pool designated for the Impala service. If YARN applications
Cloudera Administration | 247

Resource Management
are using all the original pools, and Impala uses its designated resource pool, Impala queries will have the
same resource allocation 50% x 4/8 = 25% as in the first scenario. However, when YARN applications are not
using the original pools, Impala queries will have access to 50% of the cluster resources.

The scenarios where YARN manages resources, whether for independent RM or integrated RM, map to the YARN
scheduler configuration. The scenarios where Impala independently manages resources employ the Impala
admission control feature.
To submit a YARN application to a specific resource pool, specify the mapreduce.job.queuename property. The
YARN application's queue property is mapped to a resource pool. To submit an Impala query to a specific resource
pool, specify the REQUEST_POOL option.
For details on how to configure specific resource management features, see the following topics:

Linux Control Groups (cgroups)


Required Role:
Cloudera Manager supports the Linux control groups (cgroups) kernel feature. With cgroups, administrators can
impose per-resource restrictions and limits on services and roles. This provides the ability to allocate resources
using cgroups to enable isolation of compute frameworks from one another. Resource allocation is implemented
by setting properties for the services and roles.
Linux Distribution Support
Cgroups are a feature of the Linux kernel, and as such, support depends on the host's Linux distribution and
version as shown in the following tables. If a distribution lacks support for a given parameter, changes to the
parameter have no effect.
Table 11: RHEL-compatible
Distribution
Red Hat Enterprise Linux,
CentOS, and Oracle Enterprise
Linux 7
Red Hat Enterprise Linux,
CentOS, and Oracle Enterprise
Linux 6

248 | Cloudera Administration

CPU Shares

I/O Weight

Memory Soft Limit Memory Hard Limit

Resource Management
Distribution

CPU Shares

I/O Weight

Memory Soft Limit Memory Hard Limit

CPU Shares

I/O Weight

Memory Soft Limit Memory Hard Limit

CPU Shares

I/O Weight

Memory Soft Limit Memory Hard Limit

CPU Shares

I/O Weight

Memory Soft Limit Memory Hard Limit

Red Hat Enterprise Linux,


CentOS, and Oracle Enterprise
Linux 5
Table 12: SLES
Distribution
SUSE Linux Enterprise Server
11
Table 13: Ubuntu
Distribution
Ubuntu 14.04 LTS
Ubuntu 12.04 LTS
Ubuntu 10.04 LTS
Table 14: Debian
Distribution
Debian 7.1
Debian 7.0
Debian 6.0
The exact level of support can be found in the Cloudera Manager Agent log file, shortly after the Agent has
started. See Viewing the Cloudera Manager Server Log to find the Agent log. In the log file, look for an entry like
this:
Found cgroups capabilities: {
'has_memory': True,
'default_memory_limit_in_bytes': 9223372036854775807,
'writable_cgroup_dot_procs': True,
'has_cpu': True,
'default_blkio_weight': 1000,
'default_cpu_shares': 1024,
'has_blkio': True}

The has_cpu and similar entries correspond directly to support for the CPU, I/O, and memory parameters.
Further Reading

http://www.kernel.org/doc/Documentation/cgroups/cgroups.txt
http://www.kernel.org/doc/Documentation/cgroups/blkio-controller.txt
http://www.kernel.org/doc/Documentation/cgroups/memory.txt
MANAGING SYSTEM RESOURCES ON RED HAT ENTERPRISE LINUX 6
MANAGING SYSTEM RESOURCES ON RED HAT ENTERPRISE LINUX 7

Cloudera Administration | 249

Resource Management
Resource Management with Control Groups
In order to use cgroups, you must enable cgroup-based resource management under the host resource
management configuration properties. However, if you configure static service pools, this property is set as part
of that process.
Enabling Resource Management
Cgroups-based resource management can be enabled for all hosts, or on a per-host basis.
1. If you have upgraded from a version of Cloudera Manager older than Cloudera Manager 4.5, restart every
Cloudera Manager Agent before using cgroups-based resource management:
a. Stop all services, including the Cloudera Management Service.
b. On each cluster host, run as root:
RHEL-compatible 7 and higher:
$ sudo service cloudera-scm-agent next_stop_hard
$ sudo service cloudera-scm-agent restart

All other Linux distributions:


$ sudo service cloudera-scm-agent hard_restart

c. Start all services.


2.
3.
4.
5.
6.
7.

Click the Hosts tab.


Optionally click the link of the host where you want to enable cgroups.
Click the Configuration tab.
Select Category > Resource Management.
Select Enable Cgroup-based Resource Management.
Restart all roles on the host or hosts.

Limitations
Role group and role instance override cgroup-based resource management parameters must be saved one
at a time. Otherwise some of the changes that should be reflected dynamically will be ignored.
The role group abstraction is an imperfect fit for resource management parameters, where the goal is often
to take a numeric value for a host resource and distribute it amongst running roles. The role group represents
a "horizontal" slice: the same role across a set of hosts. However, the cluster is often viewed in terms of
"vertical" slices, each being a combination of worker roles (such as TaskTracker, DataNode, RegionServer,
Impala Daemon, and so on). Nothing in Cloudera Manager guarantees that these disparate horizontal slices
are "aligned" (meaning, that the role assignment is identical across hosts). If they are unaligned, some of the
role group values will be incorrect on unaligned hosts. For example a host whose role groups have been
configured with memory limits but that's missing a role will probably have unassigned memory.

Configuring Resource Parameters


After enabling cgroups, you can restrict and limit the resource consumption of roles (or role groups) on a
per-resource basis. All of these parameters can be found in the Cloudera Manager Admin Console, under the
Resource Management category:
CPU Shares - The more CPU shares given to a role, the larger its share of the CPU when under contention.
Until processes on the host (including both roles managed by Cloudera Manager and other system processes)
are contending for all of the CPUs, this will have no effect. When there is contention, those processes with
higher CPU shares will be given more CPU time. The effect is linear: a process with 4 CPU shares will be given
roughly twice as much CPU time as a process with 2 CPU shares.

250 | Cloudera Administration

Resource Management
Updates to this parameter will be dynamically reflected in the running role.
I/O Weight - The greater the I/O weight, the higher priority will be given to I/O requests made by the role
when I/O is under contention (either by roles managed by Cloudera Manager or by other system processes).
This only affects read requests; write requests remain unprioritized. The Linux I/O scheduler controls when
buffered writes are flushed to disk, based on time and quantity thresholds. It continually flushes buffered
writes from multiple sources, not certain prioritized processes.
Updates to this parameter will be dynamically reflected in the running role.
Memory Soft Limit - When the limit is reached, the kernel will reclaim pages charged to the process if and
only if the host is facing memory pressure. If reclaiming fails, the kernel may kill the process. Both anonymous
as well as page cache pages contribute to the limit.
After updating this parameter, the role must be restarted before changes take effect.
Memory Hard Limit - When a role's resident set size (RSS) exceeds the value of this parameter, the kernel
will swap out some of the role's memory. If it's unable to do so, it will kill the process. Note that the kernel
measures memory consumption in a manner that doesn't necessarily match what the top or ps report for
RSS, so expect that this limit is a rough approximation.
After updating this parameter, the role must be restarted before changes take effect.
Example: Protecting Production MapReduce Jobs from Impala Queries
Suppose you have MapReduce deployed in production and want to roll out Impala without affecting production
MapReduce jobs. For simplicity, we will make the following assumptions:

The cluster is using homogenous hardware


Each worker host has two cores
Each worker host has 8 GB of RAM
Each worker host is running a DataNode, TaskTracker, and an Impala Daemon
Each role type is in a single role group
Cgroups-based resource management has been enabled on all hosts

Action

Procedure

CPU

1. Leave DataNode and TaskTracker role group CPU shares at 1024.


2. Set Impala Daemon role group's CPU shares to 256.
3. The TaskTracker role group should be configured with a Maximum Number of
Simultaneous Map Tasks of 2 and a Maximum Number of Simultaneous Reduce Tasks
of 1. This yields an upper bound of three MapReduce tasks at any given time; this is an
important detail for memory sizing.

Memory

1.
2.
3.
4.
5.

I/O

1. Leave DataNode and TaskTracker role group I/O weight at 500.


2. Impala Daemon role group I/O weight is set to 125.

Set Impala Daemon role group memory limit to 1024 MB.


Leave DataNode maximum Java heap size at 1 GB.
Leave TaskTracker maximum Java heap size at 1 GB.
Leave MapReduce Child Java Maximum Heap Size for Gateway at 1 GB.
Leave cgroups hard memory limits alone. We'll rely on "cooperative" memory limits
exclusively, as they yield a nicer user experience than the cgroups-based hard memory
limits.

When you're done with configuration, restart all services for these changes to take effect. The results are:
1. When MapReduce jobs are running, all Impala queries together will consume up to a fifth of the cluster's
CPU resources.
Cloudera Administration | 251

Resource Management
2. Individual Impala Daemons won't consume more than 1 GB of RAM. If this figure is exceeded, new queries
will be cancelled.
3. DataNodes and TaskTrackers can consume up to 1 GB of RAM each.
4. We expect up to 3 MapReduce tasks at a given time, each with a maximum heap size of 1 GB of RAM. That's
up to 3 GB for MapReduce tasks.
5. The remainder of each host's available RAM (6 GB) is reserved for other host processes.
6. When MapReduce jobs are running, read requests issued by Impala queries will receive a fifth of the priority
of either HDFS read requests or MapReduce read requests.

Static Service Pools


Static service pools isolate the services in your cluster from one another, so that load on one service has a
bounded impact on other services. Services are allocated a static percentage of total resourcesCPU, memory,
and I/O weightwhich are not shared with other services. When you configure static service pools, Cloudera
Manager computes recommended memory, CPU, and I/O configurations for the worker roles of the services
that correspond to the percentage assigned to each service. Static service pools are implemented per role group
within a cluster, using Linux control groups (cgroups) and cooperative memory limits (for example, Java maximum
heap sizes). Static service pools can be used to control access to resources by HBase, HDFS, Impala, MapReduce,
Solr, Spark, YARN, and add-on services. Static service pools are not enabled by default.
Note:
I/O allocation only works when short-circuit reads are enabled.
I/O allocation does not handle write side I/O because cgroups in the Linux kernel do not currently
support buffered writes.

Viewing Static Service Pool Status


Select Clusters > Cluster name > Resource Management > Static Service Pools.If the cluster has a YARN service,
the Static Service Pools Status tab displays and shows whether resource management is enabled for the cluster,
and the currently configured service pools.
See Monitoring Static Service Pools for more information.
Enabling and Configuring Static Service Pools
Required Role:
1. Select Clusters > Cluster name > Resource Management > Static Service Pools.
2. Click the Configuration tab. The Step 1 of 4: Basic Allocation Setup page displays. In each field in the basic
allocation table, enter the percentage of resources to give to each service. The total must add up to 100%. In
CDH 5 clusters, if you enable integrated resource management, the Impala service shares the YARN service
pool, rather than use its own static pool. In this case, you cannot specify a percentage for the Impala service.
Click Continue to proceed.
3. Step 2: Review Changes - The allocation of resources for each resource type and role displays with the new
values as well as the values previously in effect. The values for each role are set by role group; if there is
more than one role group for a given role type (for example, for RegionServers or DataNodes) then resources
will be allocated separately for the hosts in each role group. Take note of changed settings. If you have
previously customized these settings, check these over carefully.
Click the to the right of each percentage to display the allocations for a single service. Click
right of the Total (100%) to view all the allocations in a single page.
Click the Back button to go to the previous page and change your allocations.
When you are satisfied with the allocations, click Continue.
252 | Cloudera Administration

to the

Resource Management
4. Step 3 of 4: Restart Services - To apply the new allocation percentages, click Restart Now to restart the
cluster. To skip this step, click Restart Later. If HDFS High Availability is enabled, you will have the option to
choose a rolling restart.
5. Step 4 of 4: Progress displays the status of the restart commands. Click Finished after the restart commands
complete.
Disabling Static Service Pools
Required Role:
To disable static service pools, disable cgroup-based resource management for all hosts in all clusters:
1.
2.
3.
4.
5.
6.

In the main navigation bar, click Hosts.


Click the Configuration tab.
Select Scope > Resource Management.
Deselect the Enable Cgroup-based Resource Management property.
Click Save Changes.
Restart all services.

Static resource management is disabled, but the percentages you set when you configured the pools, and all
the changed settings (for example, heap sizes), are retained by the services. The percentages and settings will
also be used when you re-enable static service pools. If you want to revert to the settings you had before static
service pools were enabled, follow the procedures in Viewing and Reverting Configuration Changes on page 31.

Dynamic Resource Pools


Required Role:
A dynamic resource pool is a named configuration of resources and a policy for scheduling the resources among
YARN applications and Impala queries running in the pool. Dynamic resource pools allow you to schedule and
allocate resources to YARN applications and Impala queries based on a user's access to specific pools and the
resources available to those pools. If a pool's allocation is not in use it can be given to other pools. Otherwise, a
pool receives a share of resources in accordance with the pool's weight. Dynamic resource pools have ACLs that
restrict who can submit work to and administer them.
A configuration set defines the allocation of resources across pools that may be active at a given time. For
example, you can define "weekday" and "weekend" configuration sets, which define different resource pool
configurations for different days of the week.
A scheduling rule defines when a configuration set is active. The configuration set is updated in affected services
every hour.
Resource pools can be nested, with sub-pools restricted by the settings of their parent pool.
The resources available for sharing are subject to the allocations made for each service if static service pools
(cgroups) are being enforced. For example, if the static pool for YARN is 75% of the total cluster resources, then
resource pools will use only that 75% of resources.

Managing Dynamic Resource Pools


After you create or edit a resource pool,
displays while the settings are propagated to the service
configuration files. You can also manually refresh the files.

Cloudera Administration | 253

Resource Management
Viewing Dynamic Resource Pool Configuration
Depending on which resource management scenario described in Cloudera Manager Resource Management
Features on page 246 is in effect, the dynamic resource pool configuration overview displays the following
information:
YARN Independent RM - Weight, Virtual Cores, Min and Max Memory, Max Running Apps, and Scheduling
Policy
YARN and Impala Integrated RM
YARN - Weight, Virtual Cores, Min and Max Memory, Max Running Apps, and Scheduling Policy
Impala - Max Running Queries and Max Queued Queries
YARN and Impala Independent RM
YARN - Weight, Virtual Cores, Min and Max Memory, Max Running Apps, and Scheduling Policy
Impala - Max Memory, Max Running Queries, and Max Queued Queries
Impala Independent RM - Max Memory, Max Running Queries, and Max Queued Queries
To view dynamic resource pool configuration:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
Enabling and Disabling Dynamic Resource Pools for Impala
By default dynamic resource pools for Impala are disabled. If dynamic resource pools are disabled, the Impala
section will not appear in the Dynamic Resource Pools tab or in the resource pool dialogs within that page. To
modify the Impala dynamic resource pool setting:
1.
2.
3.
4.
5.
6.

Go to the Impala service.


Click the Configuration tab.
Select Category > Admission Control.
Select or deselect the Enable Dynamic Resource Pools checkbox.
Click Save Changes to commit the changes.
Restart the Impala service.

Creating a Dynamic Resource Pool


1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Add Resource Pool. The Add dialog box displays showing the General tab.
4. Specify a name and resource limits for the pool:
In the Resource Pool Name field, type a unique name containing only alphanumeric characters.
Specify the policy for scheduling resources among applications running in the pool:
Dominant Resource Fairness (DRF) (default) - An extension of fair scheduling for more than one
resourceit determines resource shares (CPU, memory) for a job separately based on the availability
of those resources and the needs of the job.
Fair Scheduler (FAIR) - Determines resource shares based on memory.
First-In, First-Out (FIFO) - Determines resource shares based on when the job was added.

254 | Cloudera Administration

Resource Management
If you have enabled Fair Scheduler preemption, optionally set a preemption timeout to specify how long
a job in this pool must wait before it can preempt resources from jobs in other pools. To enable preemption,
click the Fair Scheduler Preemption link or follow the procedure in Enabling Preemption on page 257.
5. Do one or more of the following:
Click the YARN tab.
1. Click a configuration set.
2. Specify a weight that indicates that pool's share of resources relative to other pools, minimum and
maximums for virtual cores and memory, and a limit on the number of applications that can run
simultaneously in the pool.
Click the Impala tab.
1. Click a configuration set.
2. Specify the maximum number of concurrently running and queued queries in the pool.
6. If you have enabled ACLs and specified users or groups, optionally click the Submission and Administration
Access Control tabs to specify which users and groups can submit applications and which users can view all
and kill applications. The default is that anyone can submit, view all, and kill applications. To restrict either
of these permissions, select the Allow these users and groups radio button and provide a comma-delimited
list of users and groups in the Users and Groups fields respectively. Click OK.
Adding Sub-Pools
Pools can be nested as sub-pools. They share among their siblings the resources of the parent pool. Each
sub-pool can have its own resource restrictions; if those restrictions fall within the configuration of the parent
pool, then the limits for the sub-pool take effect. If the limits for the sub-pool exceed those of the parent, then
the parent limits take effect.
Once you create sub-pools, jobs cannot be submitted to the parent pool; they must be submitted to a sub-pool.
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click at the right of a resource pool row and select Add Sub Pool. Configure sub-pool properties.
4. Click OK.
Configuring Default Scheduler Properties
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Default Settings button.
4. Specify the default scheduling policy, maximum applications, and preemption timeout properties.
5. Click OK.
Editing Dynamic Resource Pools
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
Cloudera Administration | 255

Resource Management
3. Click Edit at the right of a resource pool row. Edit the properties and click OK.
4. If you have enabled ACLs and specified users or groups, optionally click the Submission and Administration
Access Control tabs to specify which users and groups can submit applications and which users can view all
and kill applications. The default is that anyone can submit, view all, and kill applications. To restrict either
of these permissions, select the Allow these users and groups radio button and provide a comma-delimited
list of users and groups in the Users and Groups fields respectively. Click OK.
Refreshing Dynamic Resource Pool Configuration Files
After updating resource pool settings, you can refresh service configuration files as follows:
1. On the Home > Status tab, select Clusters > Cluster name > Refresh Dynamic Resource Pools.

YARN Pool Status and Configuration Options


Viewing Dynamic Resource Pool Status
Select Clusters > ClusterName > Dynamic Resource Pools. The Status tab displays the YARN resource pools
currently in use for the cluster. See Monitoring Dynamic Resource Pools for more information.
Setting User Limits
Pool properties determine the maximum number of applications that can run in a pool. To limit the number of
applications specific users can run at the same time in a pool:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
2. Click the Configuration tab.
3. Click the User Limits tab. The table displays a list of users and the maximum number of jobs each user can
submit.
4. Click Add User Limit.
5. Specify a username and maximum number of running applications.
6. Click OK.
Enabling ACLs
To specify whether ACLs are checked:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
2. Click the Configuration tab.
3. Click Other Settings.
4. In the Enable ResourceManager ACLs property, click . The YARN service configuration page displays.
5. Select the checkbox.
6. Click Save Changes to commit the changes.
7. Click to invoke the cluster restart wizard.
8. Click Restart Cluster.
9. Click Restart Now.
10. Click Finish.
Configuring ACLs
To configure which users and groups can submit and kill YARN applications in any resource pool:
1.
2.
3.
4.
5.

Enable ACLs.
Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
Click the Configuration tab.
Click Other Settings.
In the Admin ACL property, click . The YARN service configuration page displays.

256 | Cloudera Administration

Resource Management
6. Specify which users and groups can submit and kill applications.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.
Enabling Preemption
You can enable the Fair Scheduler to preempt applications in other pools if a pool's minimum share is not met
for some period of time. When you create a pool you can specify how long a pool must wait before other
applications are preempted.
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools.
2. Click the Configuration tab.
3. Click the User Limits tab. The table shows you a list of users and the maximum number of jobs each user
can submit.
4. Click Other Settings.
5. In the Fair Scheduler Preemption, click . The YARN service configuration page displays.
6. Select the checkbox.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.

Assigning Applications and Queries to Resource Pools


To submit a YARN application to a specific resource pool, specify the mapreduce.job.queuename property. The
YARN application's queue property is mapped to a resource pool. To submit an Impala query to a specific resource
pool, specify the REQUEST_POOL option.
If a specific pool is not requested, Cloudera Manager provides many options for determining how YARN applications
and Impala queries are placed in resource pools. You can specify basic rules that place applications and queries
in pools based on runtime configurations or the name of the user running the application or query or select an
advanced option that allows you to specify a set of ordered rules for placing applications and queries in pools.
Enabling and Disabling Undeclared Pools
If you do not specify a pool with a job or query property, by default YARN and Impala create a pool "on-the-fly"
with the name of the user that submitted the request and assigns it to that resource pool. For YARN, you can
change this behavior so that the default pool is used instead:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Placement Rules tab.
4. Click Basic radio button.
5. Click the Allow Undeclared Pools property.
6. Select or deselect the Allow Undeclared Pools checkbox.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.

Cloudera Administration | 257

Resource Management
10. Click Restart Now.
11. Click Finish.
Note: YARN and Impala pools created "on-the-fly" are deleted when you restart the YARN and Impala
services.
Enabling and Disabling the Default Pool
If an application specifies a pool that has not been explicitly configured or is assigned to a pool with the name
of user according to the Fair Scheduler User As Default Queue property, by default YARN creates the pool at
runtime with default settings. To change the behavior so that under these circumstances the default pool is
used instead:
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Placement Rules tab.
4. Click Basic radio button.
5. Click the Fair Scheduler User As Default Queue property.
6. Select or deselect the checkbox.
7. Click Save Changes to commit the changes.
8. Click to invoke the cluster restart wizard.
9. Click Restart Cluster.
10. Click Restart Now.
11. Click Finish.
Specifying Advanced Placement Rules and Rule Ordering
You use placement rules to indicate whether applications are placed in specified pools, pools named by a user
or group, or the default pool. To configure and order a set of rules:
1. Select the Advanced radio button on the Placement Rules tab.
2. Click to add a new rule row and to remove a rule row.
3. In each row, click and select a rule. The available rules are:
specified pool; create the pool if it doesn't exist (default 1st)
root.<username> pool; create the pool if it doesn't exist (default 2nd) - the application or query is placed
into a pool with the name of the user who submitted it.
specified pool only if the poll exists
root.<username> pool only if the pool exists
root.<primaryGroupName> pool; create the pool if it doesn't exist - the application or query is placed into
a pool with the name of the primary group of the user who submitted it.
root.<primaryGroupName> pool only if the pool exists
root.<secondaryGroupName> pool only if one of these pools exists - the application or query is placed
into a pool with a name that matches a secondary group of the user who submitted it.
default pool; create the pool if it doesn't exist
For more information about these rules, see the description of the queuePlacementPolicy element in
Allocation File Format. Reorder rules by selecting different rules for existing rule rows. If a rule is always
satisfied, subsequent rules are not evaluated and appear disabled.
4. Click Save. The Fair Scheduler allocation file (by default, fair-scheduler.xml) is updated.

258 | Cloudera Administration

Resource Management
Configuration Sets
A configuration set defines the allocation of resources across pools that may be active at a given time. For
example, you can define "weekday" and "weekend" configuration sets, which define different resource pool
configurations for different days of the week.
Creating a Configuration Set
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Scheduling Rules tab.
4. Click Add Scheduling Rule.
5. In the Configuration Set field, select the Create New radio button.
6.
7. Click Add Configuration Set. The Add Configuration Set dialog displays.
a. Type a name in the Name field and select the configuration set to clone from the Clone from Configuration
Set drop-down.
b. Click OK. The configuration set is added to and selected in the Configuration Sets drop-down.
8. For each resource pool, click Edit.
a. Select a resource pool configuration set name.
b. Edit the pool properties and click OK.
9. Define one or more scheduling rules to specify when the configuration set is active.
Example Configuration Sets
The weekday configuration set assigns the production pool four times the resources of the development pool:

The weekend configuration set assigns the production and development pools an equal share of the resources:

Cloudera Administration | 259

Resource Management
The default configuration set assigns the production pool twice the resources of the development pool:

See example scheduling rules for these configuration sets.


Viewing the Properties of a Configuration Set
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. In the Configuration Sets drop-down, select a configuration set. The properties of each pool for that
configuration set display.
Deleting a Configuration Set
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. In the Configuration Sets drop-down, select a configuration set. The properties of each pool for that
configuration set display.
4. Click Delete.

Scheduling Rules
A scheduling rule defines when a configuration set is active. The configuration set is updated in affected services
every hour.
Example Scheduling Rules
Consider the example weekday and weekend configuration sets. To specify that the weekday configuration set
is active every weekday, weekend configuration set is active on the weekend (weekly on Saturday and Sunday),
and the default configuration set is active all other times, define the following rules:

260 | Cloudera Administration

Resource Management
Adding a Scheduling Rule
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click the Scheduling Rules tab.
4. Click Add Scheduling Rule.
5. In the Configuration Set drop-down, select a configuration set.
6. Choose whether the rule should repeat, the repeat frequency, and if the frequency is weekly, the repeat day
or days.
7. If the schedule is not repeating, click the left side of the on field to display a drop-down calendar where you
set the starting date and time. When you specify the date and time, a default time window of two hours is
set in the right side of the on field. Click the right side to adjust the date and time.
8. Click OK.
Editing a Scheduling Rule
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Scheduling Rules.
4. Click Edit at the right of a rule.
5. Edit the rule as desired.
6. Click OK.
Deleting a Scheduling Rule
1. Select Clusters > Cluster name > Resource Management > Dynamic Resource Pools. If the cluster has a YARN
service, the Dynamic Resource Pools Status tab displays. If the cluster has only an Impala service enabled
for dynamic resource pools, the Dynamic Resource Pools Configuration tab displays.
2. If the Status page is displayed, click the Configuration tab. A list of the currently configured pools with their
configured limits displays.
3. Click Scheduling Rules.
4. Click at the right of a rule and select Delete.
5. Click OK.

Managing Impala Admission Control


Required Role:
Admission control is an Impala feature that imposes limits on concurrent SQL queries, to avoid resource usage
spikes and out-of-memory conditions on busy CDH clusters. It is a form of throttling. New queries are accepted
and executed until certain conditions are met, such as too many queries or too much total memory used across
the cluster. When one of these thresholds is reached, incoming queries wait to begin execution. These queries
are queued and are admitted (that is, begin executing) when the resources become available.
For further information on Impala admission control, see Admission Control and Query Queuing on page 265.
Enabling and Disabling Impala Admission Control Using Cloudera Manager
1. Go to the Impala service.
Cloudera Administration | 261

Resource Management
2.
3.
4.
5.
6.

Click the Configuration tab.


Select Category > Admission Control.
Select or deselect the Enable Impala Admission Control checkbox.
Click Save Changes to commit the changes.
Restart the Impala service.

Configuring Impala Admission Control Using Cloudera Manager


1.
2.
3.
4.

Go to the Impala service.


Click the Configuration tab.
Select Category > Admission Control.
Configure admission control properties:
default_pool_max_queued

Purpose: Maximum number of requests allowed to be queued before rejecting requests. Because this limit
applies cluster-wide, but each Impala node makes independent decisions to run queries immediately or
queue them, it is a soft limit; the overall number of queued queries might be slightly higher during times of
heavy load. A negative value or 0 indicates requests are always rejected once the maximum concurrent
requests are executing. Ignored if fair_scheduler_config_path and llama_site_path are set.
Type: int64
Default: 200
default_pool_max_requests

Purpose: Maximum number of concurrent outstanding requests allowed to run before incoming requests
are queued. Because this limit applies cluster-wide, but each Impala node makes independent decisions to
run queries immediately or queue them, it is a soft limit; the overall number of concurrent queries might be
slightly higher during times of heavy load. A negative value indicates no limit. Ignored if
fair_scheduler_config_path and llama_site_path are set.
Type: int64
Default: 200
default_pool_mem_limit

Purpose: Maximum amount of memory (across the entire cluster) that all outstanding requests in this pool
can use before new requests to this pool are queued. Specified in bytes, megabytes, or gigabytes by a number
followed by the suffix b (optional), m, or g, either uppercase or lowercase. You can specify floating-point
values for megabytes and gigabytes, to represent fractional numbers such as 1.5. You can also specify it
as a percentage of the physical memory by specifying the suffix %. 0 or no setting indicates no limit. Defaults
to bytes if no unit is given. Because this limit applies cluster-wide, but each Impala node makes independent
decisions to run queries immediately or queue them, it is a soft limit; the overall memory used by concurrent
queries might be slightly higher during times of heavy load. Ignored if fair_scheduler_config_path and
llama_site_path are set.
Note: Impala relies on the statistics produced by the COMPUTE STATS statement to estimate
memory usage for each query. See COMPUTE STATS Statement for guidelines about how and
when to use this statement.
Type: string
Default: "" (empty string, meaning unlimited)
disable_admission_control

Purpose: Turns off the admission control feature entirely, regardless of other configuration option settings.
Type: Boolean

262 | Cloudera Administration

Resource Management
Default: true
disable_pool_max_requests

Purpose: Disables all per-pool limits on the maximum number of running requests.
Type: Boolean
Default: false
disable_pool_mem_limits

Purpose: Disables all per-pool mem limits.


Type: Boolean
Default: false
fair_scheduler_allocation_path

Purpose: Path to the fair scheduler allocation file (fair-scheduler.xml).


Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as described
below. For details about all the Fair Scheduler configuration settings, see the Apache wiki.
llama_site_path

Purpose: Path to the Llama configuration file (llama-site.xml). If set, fair_scheduler_allocation_path


must also be set.
Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as described
below. For details about all the Llama configuration settings, see the documentation on Github.
queue_wait_timeout_ms

Purpose: Maximum amount of time (in milliseconds) that a request waits to be admitted before timing out.
Type: int64
Default: 60000
5. Click Save Changes to commit the changes.
6. Restart the Impala service.

Managing the Impala Llama ApplicationMaster


Note: Though Impala can be used together with YARN via simple configuration of Static Service Pools
in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.
The Impala Llama ApplicationMaster (Llama) reserves and releases YARN-managed resources for Impala, thus
reducing resource management overhead when performing Impala queries. Llama is used when you want to
enable integrated resource management.
By default, YARN allocates resources bit-by-bit as needed by MapReduce jobs. Impala needs all resources
available at the same time, so that intermediate results can be exchanged between cluster nodes, and queries
do not stall partway through waiting for new resources to be allocated. Llama is the intermediary process that
ensures all requested resources are available before each Impala query actually begins.

Cloudera Administration | 263

Resource Management
For more information about Llama, see Llama - Low Latency Application MAster.
For information on enabling Llama high availability, see Llama High Availability on page 349.

Enabling Integrated Resource Management Using Cloudera Manager


Required Role:
The Enable Integrated Resource Management wizard enables cgroups for the all the hosts in the cluster running
Impala and YARN, adds one or more Llama roles to the Impala service, and configures the Impala and YARN
services.
1. Start the wizard using one of the following paths:
Cluster-level
1. Select Clusters > Cluster name > Dynamic Resource Pools.
2. In the Status section, click Enable.
Service-level
1. Go to the Impala service.
2. Select Actions > Enable Integrated Resource Management.
The Enable Integrated Resource Management wizard starts and displays information about resource
management options and the actions performed by the wizard.
2. Click Continue.
3. Leave the Enable Cgroup-based Resource Management checkbox checked and click Continue.
4. Click the Impala Llama ApplicationMaster Hosts field to display a dialog for choosing Llama hosts.
The following shortcuts for specifying hostname patterns are supported:
Range of hostnames (without the domain portion)
Range Definition

Matching Hosts

10.1.1.[1-4]

10.1.1.1, 10.1.1.2, 10.1.1.3, 10.1.1.4

host[1-3].company.com

host1.company.com, host2.company.com, host3.company.com

host[07-10].company.com

host07.company.com, host08.company.com, host09.company.com,


host10.company.com

IP addresses
Rack name
5.
6.
7.
8.

Specify or select one or more hosts and click OK.


Click Continue. A progress screen displays with a summary of the wizard actions.
Click Continue.
Click Restart Now to restart the cluster and apply the configuration changes or click leave this wizard to
restart at a later time.
9. Click Finish.

Disabling Integrated Resource Management Using Cloudera Manager


Required Role:
The Enable Integrated Resource Management wizard enables cgroups for the all the hosts in the cluster running
Impala and YARN, adds one or more Llama roles to the Impala service, and configures the Impala and YARN
services.

264 | Cloudera Administration

Resource Management
1. Start the wizard using one of the following paths:
Cluster-level
1. Select Clusters > Cluster name > Dynamic Resource Pools.
2. In the Status section, click Disable.
Service-level
1. Go to the Impala service.
2. Select Actions > Disable Integrated Resource Management.
The Disable Integrated Resource Management wizard starts and displays information about resource
management options and the actions performed by the wizard.

Configuring Llama Using Cloudera Manager


Required Role:
1.
2.
3.
4.
5.
6.

Go to the Impala service.


Click the Configuration tab.
Select Scope > Impala Llama ApplicationMaster.
Edit configuration properties.
Click Save Changes to commit the changes.
Restart the Llama role.

Impala Resource Management


Impala supports two types of resource management: independent and integrated. Independent resource
management is supported for CDH 4 and CDH 5 and is implemented by admission control. Integrated resource
management is supported for CDH 5 and is implemented by YARN and Llama.
Note: Though Impala can be used together with YARN via simple configuration of Static Service Pools
in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.

Admission Control and Query Queuing


Admission control is an Impala feature that imposes limits on concurrent SQL queries, to avoid resource usage
spikes and out-of-memory conditions on busy CDH clusters. It is a form of throttling. New queries are accepted
and executed until certain conditions are met, such as too many queries or too much total memory used across
the cluster. When one of these thresholds is reached, incoming queries wait to begin execution. These queries
are queued and are admitted (that is, begin executing) when the resources become available.
In addition to the threshold values for currently executing queries, you can place limits on the maximum number
of queries that are queued (waiting) and a limit on the amount of time they might wait before returning with an
error. These queue settings let you ensure that queries do not wait indefinitely, so that you can detect and
correct starvation scenarios.
Enable this feature if your cluster is underutilized at some times and overutilized at others. Overutilization is
indicated by performance bottlenecks and queries being cancelled due to out-of-memory conditions, when
those same queries are successful and perform well during times with less concurrent load. Admission control
works as a safeguard to avoid out-of-memory conditions during heavy concurrent usage.

Cloudera Administration | 265

Resource Management
Important:
Cloudera strongly recommends you upgrade to CDH 5 or higher to use admission control. In CDH
4, admission control will only work if you don't have Hue deployed; unclosed Hue queries will
accumulate and exceed the queue size limit. On CDH 4, to use admission control, you must explicitly
enable it by specifying --disable_admission_control=false in the impalad command-line
options.
Use the COMPUTE STATS statement for large tables involved in join queries, and follow other steps
from Tuning Impala for Performance to tune your queries. Although COMPUTE STATS is an important
statement to help optimize query performance, it is especially important when admission control
is enabled:
When queries complete quickly and are tuned for optimal memory usage, there is less chance
of performance or capacity problems during times of heavy load.
The admission control feature also relies on the statistics produced by the COMPUTE STATS
statement to generate accurate estimates of memory usage for complex queries. If the
estimates are inaccurate due to missing statistics, Impala might hold back queries unnecessarily
even though there is sufficient memory to run them, or might allow queries to run that end
up exceeding the memory limit and being cancelled.

Overview of Impala Admission Control


On a busy CDH cluster, you might find there is an optimal number of Impala queries that run concurrently.
Because Impala queries are typically I/O-intensive, you might not find any throughput benefit in running more
concurrent queries when the I/O capacity is fully utilized. Because Impala by default cancels queries that exceed
the specified memory limit, running multiple large-scale queries at once can result in having to re-run some
queries that are cancelled.
The admission control feature lets you set a cluster-wide upper limit on the number of concurrent Impala queries
and on the memory used by those queries. Any additional queries are queued until the earlier ones finish, rather
than being cancelled or running slowly and causing contention. As other queries finish, the queued queries are
allowed to proceed.
For details on the internal workings of admission control, see How Impala Schedules and Enforces Limits on
Concurrent Queries on page 267.

How Impala Admission Control Relates to YARN


The admission control feature is similar in some ways to the YARN resource management framework, and they
can be used separately or together. This section describes some similarities and differences, to help you decide
when to use one, the other, or both together.
Admission control is a lightweight, decentralized system that is suitable for workloads consisting primarily of
Impala queries and other SQL statements. It sets soft limits that smooth out Impala memory usage during
times of heavy load, rather than taking an all-or-nothing approach that cancels jobs that are too
resource-intensive.
Because the admission control system is not aware of other Hadoop workloads such as MapReduce jobs, you
might use YARN with static service pools on heterogeneous CDH 5 clusters where resources are shared between
Impala and other Hadoop components. Devote a percentage of cluster resources to Impala, allocate another
percentage for MapReduce and other batch-style workloads; let admission control handle the concurrency and
memory usage for the Impala work within the cluster, and let YARN manage the remainder of work within the
cluster.
The Impala admission control feature uses the same mechanism as the YARN resource manager to map users
to pools and authenticate them. Although the YARN resource manager is only available with CDH 5 and higher,
internally Impala includes the necessary infrastructure to work consistently on both CDH 4 and CDH 5. You do
not need to run YARN for admission control to operate.

266 | Cloudera Administration

Resource Management
For full details about using Impala with YARN, see Integrated Resource Management with YARN on page 273.

How Impala Schedules and Enforces Limits on Concurrent Queries


The admission control system is decentralized, embedded in each Impala daemon and communicating through
the statestore mechanism. Although the limits you set for memory usage and number of concurrent queries
apply cluster-wide, each Impala daemon makes its own decisions about whether to allow each query to run
immediately or to queue it for a less-busy time. These decisions are fast, meaning the admission control
mechanism is low-overhead, but might be imprecise during times of heavy load. There could be times when the
query queue contained more queries than the specified limit, or when the estimated of memory usage for a
query is not exact and the overall memory usage exceeds the specified limit. Thus, you typically err on the high
side for the size of the queue, because there is not a big penalty for having a large number of queued queries;
and you typically err on the low side for the memory limit, to leave some headroom for queries to use more
memory than expected, without being cancelled as a result.
At any time, the set of queued queries could include queries submitted through multiple different Impala daemon
hosts. All the queries submitted through a particular host will be executed in order, so a CREATE TABLE followed
by an INSERT on the same table would succeed. Queries submitted through different hosts are not guaranteed
to be executed in the order they were received. Therefore, if you are using load-balancing or other round-robin
scheduling where different statements are submitted through different hosts, set up all table structures ahead
of time so that the statements controlled by the queuing system are primarily queries, where order is not
significant. Or, if a sequence of statements needs to happen in strict order (such as an INSERT followed by a
SELECT), submit all those statements through a single session, while connected to the same Impala daemon
host.
The limit on the number of concurrent queries is a soft one, To achieve high throughput, Impala makes quick
decisions at the host level about which queued queries to dispatch. Therefore, Impala might slightly exceed the
limit from time to time.
To avoid a large backlog of queued requests, you can also set an upper limit on the size of the queue for queries
that are delayed. When the number of queued queries exceeds this limit, further queries are cancelled rather
than being queued. You can also configure a timeout period, after which queued queries are cancelled, to avoid
indefinite waits. If a cluster reaches this state where queries are cancelled due to too many concurrent requests
or long waits for query execution to begin, that is a signal for an administrator to take action, either by provisioning
more resources, scheduling work on the cluster to smooth out the load, or by doing Impala performance tuning
to enable higher throughput.

How Admission Control works with Impala Clients (JDBC, ODBC, HiveServer2)
Most aspects of admission control work transparently with client interfaces such as JDBC and ODBC:
If a SQL statement is put into a queue rather than running immediately, the API call blocks until the statement
is dequeued and begins execution. At that point, the client program can request to fetch results, which might
also block until results become available.
If a SQL statement is cancelled because it has been queued for too long or because it exceeded the memory
limit during execution, the error is returned to the client program with a descriptive error message.
If you want to submit queries to different resource pools through the REQUEST_POOL query option, as described
in REQUEST_POOL Query Option, In Impala 2.0 and higher you can change that query option through a SQL SET
statement that you submit from the client application, in the same session. Prior to Impala 2.0, that option was
only settable for a session through the impala-shell SET command, or cluster-wide through an impalad
startup option.
Admission control has the following limitations or special behavior when used with JDBC or ODBC applications:
The MEM_LIMIT query option, sometimes useful to work around problems caused by inaccurate memory
estimates for complicated queries, is only settable through the impala-shell interpreter and cannot be
used directly through JDBC or ODBC applications.
Admission control does not use the other resource-related query options, RESERVATION_REQUEST_TIMEOUT
or V_CPU_CORES. Those query options only apply to the YARN resource management framework.

Cloudera Administration | 267

Resource Management
Configuring Admission Control
The configuration options for admission control range from the simple (a single resource pool with a single set
of options) to the complex (multiple resource pools with different options, each pool handling queries for a
different set of users and groups). You can configure the settings through the Cloudera Manager user interface,
or on a system without Cloudera Manager by editing configuration files or through startup options to the impalad
daemon.
Admission Control Options
The following Impala configuration options let you adjust the settings of the admission control feature. When
supplying the options on the command line, prepend the option name with --.
default_pool_max_queued

Purpose: Maximum number of requests allowed to be queued before rejecting requests. Because this
limit applies cluster-wide, but each Impala node makes independent decisions to run queries immediately
or queue them, it is a soft limit; the overall number of queued queries might be slightly higher during
times of heavy load. A negative value or 0 indicates requests are always rejected once the maximum
concurrent requests are executing. Ignored if fair_scheduler_config_path and llama_site_path
are set.
Type: int64
Default: 200
default_pool_max_requests

Purpose: Maximum number of concurrent outstanding requests allowed to run before incoming requests
are queued. Because this limit applies cluster-wide, but each Impala node makes independent decisions
to run queries immediately or queue them, it is a soft limit; the overall number of concurrent queries
might be slightly higher during times of heavy load. A negative value indicates no limit. Ignored if
fair_scheduler_config_path and llama_site_path are set.
Type: int64
Default: 200
default_pool_mem_limit

Purpose: Maximum amount of memory (across the entire cluster) that all outstanding requests in this
pool can use before new requests to this pool are queued. Specified in bytes, megabytes, or gigabytes
by a number followed by the suffix b (optional), m, or g, either uppercase or lowercase. You can specify
floating-point values for megabytes and gigabytes, to represent fractional numbers such as 1.5. You
can also specify it as a percentage of the physical memory by specifying the suffix %. 0 or no setting
indicates no limit. Defaults to bytes if no unit is given. Because this limit applies cluster-wide, but each
Impala node makes independent decisions to run queries immediately or queue them, it is a soft limit;
the overall memory used by concurrent queries might be slightly higher during times of heavy load.
Ignored if fair_scheduler_config_path and llama_site_path are set.
Note: Impala relies on the statistics produced by the COMPUTE STATS statement to estimate
memory usage for each query. See COMPUTE STATS Statement for guidelines about how and
when to use this statement.
Type: string
Default: "" (empty string, meaning unlimited)
disable_admission_control

Purpose: Turns off the admission control feature entirely, regardless of other configuration option settings.
Type: Boolean

268 | Cloudera Administration

Resource Management
Default: true
disable_pool_max_requests

Purpose: Disables all per-pool limits on the maximum number of running requests.
Type: Boolean
Default: false
disable_pool_mem_limits

Purpose: Disables all per-pool mem limits.


Type: Boolean
Default: false
fair_scheduler_allocation_path

Purpose: Path to the fair scheduler allocation file (fair-scheduler.xml).


Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as
described below. For details about all the Fair Scheduler configuration settings, see the Apache wiki.
llama_site_path

Purpose: Path to the Llama configuration file (llama-site.xml). If set,


fair_scheduler_allocation_path must also be set.
Type: string
Default: "" (empty string)
Usage notes: Admission control only uses a small subset of the settings that can go in this file, as
described below. For details about all the Llama configuration settings, see the documentation on Github.
queue_wait_timeout_ms

Purpose: Maximum amount of time (in milliseconds) that a request waits to be admitted before timing
out.
Type: int64
Default: 60000
Configuring Admission Control Using Cloudera Manager
In Cloudera Manager, you can configure pools to manage queued Impala queries, and the options for the limit
on number of concurrent queries and how to handle queries that exceed the limit. For details, see Managing
Resources with Cloudera Manager.
See Examples of Admission Control Configurations on page 270 for a sample setup for admission control under
Cloudera Manager.
Configuring Admission Control Using the Command Line
If you do not use Cloudera Manager, you use a combination of startup options for the Impala daemon, and
optionally editing or manually constructing the configuration files fair-scheduler.xml and llama-site.xml.
For a straightforward configuration using a single resource pool named default, you can specify configuration
options on the command line and skip the fair-scheduler.xml and llama-site.xml configuration files.
For an advanced configuration with multiple resource pools using different settings, set up the
fair-scheduler.xml and llama-site.xml configuration files manually. Provide the paths to each one using
the Impala daemon command-line options, --fair_scheduler_allocation_path and --llama_site_path
respectively.

Cloudera Administration | 269

Resource Management
The Impala admission control feature only uses the Fair Scheduler configuration settings to determine how to
map users and groups to different resource pools. For example, you might set up different resource pools with
separate memory limits, and maximum number of concurrent and queued queries, for different categories of
users within your organization. For details about all the Fair Scheduler configuration settings, see the Apache
wiki.
The Impala admission control feature only uses a small subset of possible settings from the llama-site.xml
configuration file:
llama.am.throttling.maximum.placed.reservations.queue_name
llama.am.throttling.maximum.queued.reservations.queue_name

For details about all the Llama configuration settings, see Llama Default Configuration.
See Example Admission Control Configurations Using Configuration Files on page 271 for sample configuration
files for admission control using multiple resource pools, without Cloudera Manager.
Examples of Admission Control Configurations
Example Admission Control Configurations Using Cloudera Manager
For full instructions about configuring dynamic resource pools through Cloudera Manager, see Dynamic Resource
Pools on page 253. The following examples demonstrate some important points related to the Impala admission
control feature.
The following figure shows a sample of the Dynamic Resource Pools page in Cloudera Manager, accessed through
the Clusters > Cluster name > Resource Management > Dynamic Resource Pools menu choice and then the
Configuration tab. Numbers from all the resource pools are combined into the topmost root pool. The default
pool is for users who are not assigned any other pool by the user-to-pool mapping settings. The development
and production pools show how you can set different limits for different classes of users, for total memory,
number of concurrent queries, and number of queries that can be queued.

Figure 7: Sample Settings for Cloudera Manager Dynamic Resource Pools Page
The following figure shows a sample of the Placement Rules page in Cloudera Manager, accessed through the
Clusters > Cluster name > Resource Management > Dynamic Resource Pools menu choice and then the

270 | Cloudera Administration

Resource Management
Configuration > Placement Rules tabs. The settings demonstrate a reasonable configuration of a pool named
default to service all requests where the specified resource pool does not exist, is not explicitly set, or the user
or group is not authorized for the specified pool.

Figure 8: Sample Settings for Cloudera Manager Placement Rules Page


Example Admission Control Configurations Using Configuration Files
For clusters not managed by Cloudera Manager, here are sample fair-scheduler.xml and llama-site.xml
files that define resource pools equivalent to the ones in the preceding Cloudera Manager dialog. These sample
files are stripped down: in a real deployment they might contain other settings for use with various aspects of
the YARN and Llama components. The settings shown here are the significant ones for the Impala admission
control feature.
fair-scheduler.xml:
Although Impala does not use the vcores value, you must still specify it to satisfy YARN requirements for the
file contents.
Each <aclSubmitApps> tag (other than the one for root) contains a comma-separated list of users, then a
space, then a comma-separated list of groups; these are the users and groups allowed to submit Impala
statements to the corresponding resource pool.
If you leave the <aclSubmitApps> element empty for a pool, nobody can submit directly to that pool; child pools
can specify their own <aclSubmitApps> values to authorize users and groups to submit to those pools.
<allocations>
<queue name="root">
<aclSubmitApps> </aclSubmitApps>
<queue name="default">
<maxResources>50000 mb, 0 vcores</maxResources>
<aclSubmitApps>*</aclSubmitApps>
</queue>
<queue name="development">
<maxResources>200000 mb, 0 vcores</maxResources>
<aclSubmitApps>user1,user2 dev,ops,admin</aclSubmitApps>
</queue>
<queue name="production">
<maxResources>1000000 mb, 0 vcores</maxResources>
<aclSubmitApps> ops,admin</aclSubmitApps>

Cloudera Administration | 271

Resource Management
</queue>
</queue>
<queuePlacementPolicy>
<rule name="specified" create="false"/>
<rule name="default" />
</queuePlacementPolicy>
</allocations>

llama-site.xml:
<?xml version="1.0" encoding="UTF-8"?>
<configuration>
<property>
<name>llama.am.throttling.maximum.placed.reservations.root.default</name>
<value>10</value>
</property>
<property>
<name>llama.am.throttling.maximum.queued.reservations.root.default</name>
<value>50</value>
</property>
<property>
<name>llama.am.throttling.maximum.placed.reservations.root.development</name>
<value>50</value>
</property>
<property>
<name>llama.am.throttling.maximum.queued.reservations.root.development</name>
<value>100</value>
</property>
<property>
<name>llama.am.throttling.maximum.placed.reservations.root.production</name>
<value>100</value>
</property>
<property>
<name>llama.am.throttling.maximum.queued.reservations.root.production</name>
<value>200</value>
</property>
</configuration>

Guidelines for Using Admission Control


To see how admission control works for particular queries, examine the profile output for the query. This
information is available through the PROFILE statement in impala-shell immediately after running a query
in the shell, on the queries page of the Impala debug web UI, or in the Impala log file (basic information at log
level 1, more detailed information at log level 2). The profile output contains details about the admission decision,
such as whether the query was queued or not and which resource pool it was assigned to. It also includes the
estimated and actual memory usage for the query, so you can fine-tune the configuration for the memory limits
of the resource pools.
Where practical, use Cloudera Manager to configure the admission control parameters. The Cloudera Manager
GUI is much simpler than editing the configuration files directly.
Remember that the limits imposed by admission control are soft limits. Although the limits you specify for
number of concurrent queries and amount of memory apply cluster-wide, the decentralized nature of this
mechanism means that each Impala node makes its own decisions about whether to allow queries to run
immediately or to queue them. These decisions rely on information passed back and forth between nodes by
the statestore service. If a sudden surge in requests causes more queries than anticipated to run concurrently,
then as a fallback, the overall Impala memory limit and the Linux cgroups mechanism serve as hard limits to
prevent overallocation of memory, by cancelling queries if necessary.
If you have trouble getting a query to run because its estimated memory usage is too high, you can override the
estimate by setting the MEM_LIMIT query option in impala-shell, then issuing the query through the shell in
the same session. The MEM_LIMIT value is treated as the estimated amount of memory, overriding the estimate
that Impala would generate based on table and column statistics. This value is used only for making admission
control decisions, and is not pre-allocated by the query.

272 | Cloudera Administration

Resource Management
In impala-shell, you can also specify which resource pool to direct queries to by setting the REQUEST_POOL
query option. (This option was named YARN_POOL during the CDH 5 beta period.)
The statements affected by the admission control feature are primarily queries, but also include statements
that write data such as INSERT and CREATE TABLE AS SELECT. Most write operations in Impala are not
resource-intensive, but inserting into a Parquet table can require substantial memory due to buffering 1 GB of
data before writing out each Parquet data block. See Loading Data into Parquet Tables for instructions about
inserting data efficiently into Parquet tables.
Although admission control does not scrutinize memory usage for other kinds of DDL statements, if a query is
queued due to a limit on concurrent queries or memory usage, subsequent statements in the same session are
also queued so that they are processed in the correct order:
-- This query could be queued to avoid out-of-memory at times of heavy load.
select * from huge_table join enormous_table using (id);
-- If so, this subsequent statement in the same session is also queued
-- until the previous statement completes.
drop table huge_table;

If you set up different resource pools for different users and groups, consider reusing any classifications and
hierarchy you developed for use with Sentry security. See Enabling Sentry Authorization for Impala for details.
For details about all the Fair Scheduler configuration settings, see Fair Scheduler Configuration, in particular the
tags such as <queue> and <aclSubmitApps> to map users and groups to particular resource pools (queues).

Integrated Resource Management with YARN


You can limit the CPU and memory resources used by Impala, to manage and prioritize workloads on clusters
that run jobs from many Hadoop components. (Currently, there is no limit or throttling on the I/O for Impala
queries.) In CDH 5, Impala can use the underlying Apache Hadoop YARN resource management framework, which
allocates the required resources for each Impala query. Impala estimates the resources required by the query
on each host of the cluster, and requests the resources from YARN.
Requests from Impala to YARN go through an intermediary service called Llama. When the resource requests
are granted, Impala starts the query and places all relevant execution threads into the cgroup containers and
sets up the memory limit on each host. If sufficient resources are not available, the Impala query waits until
other jobs complete and the resources are freed. During query processing, as the need for additional resources
arises, Llama can expand already-requested resources, to avoid over-allocating at the start of the query.
After a query is finished, Llama caches the resources (for example, leaving memory allocated) in case they are
needed for subsequent Impala queries. This caching mechanism avoids the latency involved in making a whole
new set of resource requests for each query. If the resources are needed by YARN for other types of jobs, Llama
returns them.
While the delays to wait for resources might make individual queries seem less responsive on a heavily loaded
cluster, the resource management feature makes the overall performance of the cluster smoother and more
predictable, without sudden spikes in utilization due to memory paging, CPUs pegged at 100%, and so on.
Note: Though Impala can be used together with YARN via simple configuration of Static Service Pools
in Cloudera Manager, the use of the general-purpose component Llama for integrated resource
management within YARN is no longer supported with CDH 5.5 / Impala 2.3 and higher.

The Llama Daemon


Llama is a system that mediates resource management between Cloudera Impala and Hadoop YARN. Llama
enables Impala to reserve, use, and release resource allocations in a Hadoop cluster. Llama is only required if
resource management is enabled in Impala.
By default, YARN allocates resources bit-by-bit as needed by MapReduce jobs. Impala needs all resources
available at the same time, so that intermediate results can be exchanged between cluster nodes, and queries

Cloudera Administration | 273

Resource Management
do not stall partway through waiting for new resources to be allocated. Llama is the intermediary process that
ensures all requested resources are available before each Impala query actually begins.
For management through Cloudera Manager, see The Impala Llama ApplicationMaster.

Controlling Resource Estimation Behavior


By default, Impala consults the table statistics and column statistics for each table in a query, and uses those
figures to construct estimates of needed resources for each query. See COMPUTE STATS Statement for the
statement to collect table and column statistics for a table.
To avoid problems with inaccurate or missing statistics, which can lead to inaccurate estimates of resource
consumption, Impala allows you to set default estimates for CPU and memory resource consumption. As a query
grows to require more resources, Impala will request more resources from Llama (this is called expanding a
query reservation). When the query is complete, those resources are returned to YARN as normal. To enable
this feature, use the command-line option -rm_always_use_defaults when starting impalad, and optionally
-rm_default_memory=size and -rm_default_cpu_cores. Cloudera recommends always running with
-rm_always_use_defaults enabled when using resource management, because if the query needs more
resources than the default values, the resource requests are expanded dynamically as the query runs. See
impalad Startup Options for Resource Management on page 275 for details about each option.

Checking Resource Estimates and Actual Usage


To make resource usage easier to verify, the output of the EXPLAIN SQL statement now includes information
about estimated memory usage, whether table and column statistics are available for each table, and the number
of virtual cores that a query will use. You can get this information through the EXPLAIN statement without
actually running the query. The extra information requires setting the query option EXPLAIN_LEVEL=verbose;
see EXPLAIN Statement for details. The same extended information is shown at the start of the output from
the PROFILE statement in impala-shell. The detailed profile information is only available after running the
query. You can take appropriate actions (gathering statistics, adjusting query options) if you find that queries
fail or run with suboptimal performance when resource management is enabled.

How Resource Limits Are Enforced


CPU limits are enforced by the Linux cgroups mechanism. YARN grants resources in the form of containers
that correspond to cgroups on the respective machines.
Memory is enforced by Impala's query memory limits. Once a reservation request has been granted, Impala
sets the query memory limit according to the granted amount of memory before executing the query.

Enabling Resource Management for Impala


To enable resource management for Impala, first you set up the YARN and Llama services for your CDH cluster.
Then you add startup options and customize resource management settings for the Impala services.
Required CDH Setup for Resource Management with Impala
YARN is the general-purpose service that manages resources for many Hadoop components within a CDH
cluster. Llama is a specialized service that acts as an intermediary between Impala and YARN, translating Impala
resource requests to YARN and coordinating with Impala so that queries only begin executing when all needed
resources have been granted by YARN.
For information about setting up the YARN and Llama services, see the instructions for Cloudera Manager.
Using Impala with a Llama High Availability Configuration
Impala can take advantage of the Llama high availability (HA) feature, with additional Llama servers that step
in if the primary one becomes unavailable. (Only one Llama server at a time services all resource requests.)
Before using this feature from Impala, read the background information about Llama HA, its main features, and
how to set it up.
Command-line method for systems without Cloudera Manager:

274 | Cloudera Administration

Resource Management
Setting up the Impala side in a Llama HA configuration involves setting the impalad configuration options
-llama_addresses (mandatory) and optionally -llama_max_request_attempts,
-llama_registration_timeout_secs, and -llama_registration_wait_secs. See the next section impalad
Startup Options for Resource Management on page 275 for usage instructions for those options.
The impalad daemon on the coordinator host registers with the Llama server for each query, receiving a handle
that is used for subsequent resource requests. If a Llama server becomes unavailable, all running Impala queries
are cancelled. Subsequent queries register with the next specified Llama server. This registration only happens
when a query or similar request causes an impalad to request resources through Llama. Therefore, when a
Llama server becomes unavailable, that fact might not be reported immediately in the Impala status information
such as the metrics page in the debug web UI.
Cloudera Manager method: See Llama High Availailability.
impalad Startup Options for Resource Management
The following startup options for impalad enable resource management and customize its parameters for your
cluster configuration:
-enable_rm: Whether to enable resource management or not, either true or false. The default is false.
None of the other resource management options have any effect unless -enable_rm is turned on.
-llama_host: Hostname or IP address of the Llama service that Impala should connect to. The default is
127.0.0.1.
-llama_port: Port of the Llama service that Impala should connect to. The default is 15000.
-llama_callback_port: Port that Impala should start its Llama callback service on. Llama reports when
resources are granted or preempted through that service.
-cgroup_hierarchy_path: Path where YARN and Llama will create cgroups for granted resources. Impala
assumes that the cgroup for an allocated container is created in the path 'cgroup_hierarchy_path +
container_id'.
-rm_always_use_defaults: If this Boolean option is enabled, Impala ignores computed estimates and
always obtains the default memory and CPU allocation from Llama at the start of the query. These default
estimates are approximately 2 CPUs and 4 GB of memory, possibly varying slightly depending on cluster size,
workload, and so on. Cloudera recommends enabling -rm_always_use_defaults whenever resource
management is used, and relying on these default values (that is, leaving out the two following options).
-rm_default_memory=size: Optionally sets the default estimate for memory usage for each query. You
can use suffixes such as M and G for megabytes and gigabytes, the same as with the MEM_LIMIT query
option. Only has an effect when -rm_always_use_defaults is also enabled.
-rm_default_cpu_cores: Optionally sets the default estimate for number of virtual CPU cores for each
query. Only has an effect when -rm_always_use_defaults is also enabled.
The following options fine-tune the interaction of Impala with Llama when Llama high availability (HA) is enabled.
The -llama_addresses option is only applicable in a Llama HA environment. -llama_max_request_attempts,
-llama_registration_timeout_secs, and -llama_registration_wait_secs work whether or not Llama
HA is enabled, but are most useful in combination when Llama is set up for high availability.
-llama_addresses: Comma-separated list of hostname:port items, specifying all the members of the
Llama availability group. Defaults to "127.0.0.1:15000".
-llama_max_request_attempts: Maximum number of times a request to reserve, expand, or release
resources is retried until the request is cancelled. Attempts are only counted after Impala is registered with
Llama. That is, a request survives at mostllama_max_request_attempts-1 re-registrations. Defaults to
5.
-llama_registration_timeout_secs: Maximum number of seconds that Impala will attempt to register
or re-register with Llama. If registration is unsuccessful, Impala cancels the action with an error, which could
result in an impalad startup failure or a cancelled query. A setting of -1 means try indefinitely. Defaults to
30.
-llama_registration_wait_secs: Number of seconds to wait between attempts during Llama registration.
Defaults to 3.

Cloudera Administration | 275

Resource Management
impala-shell Query Options for Resource Management
Before issuing SQL statements through the impala-shell interpreter, you can use the SET command to configure
the following parameters related to resource management:

EXPLAIN_LEVEL Query Option


MEM_LIMIT Query Option
RESERVATION_REQUEST_TIMEOUT Query Option (CDH 5 only)
V_CPU_CORES Query Option (CDH 5 only)

Limitations of Resource Management for Impala


Currently, Impala in CDH 5 has the following limitations for resource management of Impala queries:
Table statistics are required, and column statistics are highly valuable, for Impala to produce accurate
estimates of how much memory to request from YARN. See Overview of Table Statistics and Overview of
Column Statistics for instructions on gathering both kinds of statistics, and EXPLAIN Statement for the
extended EXPLAIN output where you can check that statistics are available for a specific table and set of
columns.
If the Impala estimate of required memory is lower than is actually required for a query, Impala dynamically
expands the amount of requested memory. Queries might still be cancelled if the reservation expansion fails,
for example if there are insufficient remaining resources for that pool, or the expansion request takes long
enough that it exceeds the query timeout interval, or because of YARN preemption. You can see the actual
memory usage after a failed query by issuing a PROFILE command in impala-shell. Specify a larger memory
figure with the MEM_LIMIT query option and re-try the query.
The MEM_LIMIT query option, and the other resource-related query options, are settable through the ODBC or
JDBC interfaces in Impala 2.0 and higher. This is a former limitation that is now lifted.

276 | Cloudera Administration

Performance Management

Performance Management
This section describes mechanisms and best practices for improving performance.
Related Information
Tuning Impala for Performance
Tuning the Cluster for MapReduce v2 (YARN)

Optimizing Performance in CDH


This section provides solutions to some performance problems, and describes configuration best practices.
Important: If you are running CDH over 10 GBps Ethernet, improperly set network configuration or
improperly applied NIC firmware or drivers can noticeably degrade performance. Work with your
network engineers and hardware vendors to make sure that you have the proper NIC firmware, drivers,
and configurations in place and that your network performs properly. Cloudera recognizes that network
setup and upgrade are challenging problems, and will make best efforts to share any helpful
experiences.
Disabling Transparent Hugepage Compaction
Most Linux platforms supported by CDH 5 include a feature called transparent hugepage compaction which
interacts poorly with Hadoop workloads and can seriously degrade performance.
Symptom: top and other system monitoring tools show a large percentage of the CPU usage classified as "system
CPU". If system CPU usage is 30% or more of the total CPU usage, your system may be experiencing this issue.
What to do:
Note: In the following instructions, defrag_file_pathname depends on your operating system:
Red Hat/CentOS: /sys/kernel/mm/redhat_transparent_hugepage/defrag
Ubuntu/Debian, OEL, SLES: /sys/kernel/mm/transparent_hugepage/defrag
1. To see whether transparent hugepage compaction is enabled, run the following command and check the
output:
$ cat defrag_file_pathname

[always] never means that transparent hugepage compaction is enabled.


always [never] means that transparent hugepage compaction is disabled.
2. To disable transparent hugepage compaction, add the following command to /etc/rc.local:
echo never > defrag_file_pathname

You can also disable transparent hugepage compaction interactively (but remember this will not survive a
reboot).
To disable transparent hugepage compaction temporarily as root:
# echo 'never' > defrag_file_pathname

Cloudera Administration | 277

Performance Management
To disable transparent hugepage compaction temporarily using sudo:
$ sudo sh -c "echo 'never' > defrag_file_pathname"

Setting the vm.swappiness Linux Kernel Parameter


vm.swappiness is a Linux kernel parameter that controls how aggressively memory pages are swapped to disk.

It can be set to a value between 0-100; the higher the value, the more aggressive the kernel is in seeking out
inactive memory pages and swapping them to disk.
You can see what value vm.swappiness is currently set to by looking at /proc/sys/vm; for example:
cat /proc/sys/vm/swappiness

On most systems, it is set to 60 by default. This is not suitable for Hadoop cluster nodes, because it can cause
processes to get swapped out even when there is free memory available. This can affect stability and performance,
and may cause problems such as lengthy garbage collection pauses for important system daemons. Cloudera
recommends that you set this parameter to 10 or less; for example:
# sysctl -w vm.swappiness=10

Cloudera previously recommended a setting of 0, but in recent kernels (such as those included with RedHat 6.4
and higher, and Ubuntu 12.04 LTS and higher) a setting of 0 might lead to out of memory issues per this blog
post: http://www.percona.com/blog/2014/04/28/oom-relation-vm-swappiness0-new-kernel/.
Improving Performance in Shuffle Handler and IFile Reader
The MapReduce shuffle handler and IFile reader use native Linux calls (posix_fadvise(2) and sync_data_range)
on Linux systems with Hadoop native libraries installed. The subsections that follow provide details.
Shuffle Handler
You can improve MapReduce shuffle handler performance by enabling shuffle readahead. This causes the
TaskTracker or Node Manager to pre-fetch map output before sending it over the socket to the reducer.
To enable this feature for YARN, set the mapreduce.shuffle.manage.os.cache property to true (default).
To further tune performance, adjust the value of the mapreduce.shuffle.readahead.bytes property. The
default value is 4MB.
To enable this feature for MRv1, set the mapred.tasktracker.shuffle.fadvise property to true (default).
To further tune performance, adjust the value of the mapred.tasktracker.shuffle.readahead.bytes
property. The default value is 4MB.
IFile Reader
Enabling IFile readahead increases the performance of merge operations. To enable this feature for either MRv1
or YARN, set the mapreduce.ifile.readahead property to true (default). To further tune the performance,
adjust the value of the mapreduce.ifile.readahead.bytes property. The default value is 4MB.
Best Practices for MapReduce Configuration
The configuration settings described below can reduce inherent latencies in MapReduce execution. You set
these values in mapred-site.xml.
Send a heartbeat as soon as a task finishes
Set the mapreduce.tasktracker.outofband.heartbeat property to true to let the TaskTracker send an
out-of-band heartbeat on task completion to reduce latency; the default value is false:
<property>
<name>mapreduce.tasktracker.outofband.heartbeat</name>

278 | Cloudera Administration

Performance Management
<value>true</value>
</property>

Reduce the interval for JobClient status reports on single node systems
The jobclient.progress.monitor.poll.interval property defines the interval (in milliseconds) at which
JobClient reports status to the console and checks for job completion. The default value is 1000 milliseconds;
you may want to set this to a lower value to make tests run faster on a single-node cluster. Adjusting this value
on a large production cluster may lead to unwanted client-server traffic.
<property>
<name>jobclient.progress.monitor.poll.interval</name>
<value>10</value>
</property>

Tune the JobTracker heartbeat interval


Tuning the minimum interval for the TaskTracker-to-JobTracker heartbeat to a smaller value may improve
MapReduce performance on small clusters.
<property>
<name>mapreduce.jobtracker.heartbeat.interval.min</name>
<value>10</value>
</property>

Start MapReduce JVMs immediately


The mapred.reduce.slowstart.completed.maps property specifies the proportion of Map tasks in a job that
must be completed before any Reduce tasks are scheduled. For small jobs that require fast turnaround, setting
this value to 0 can improve performance; larger values (as high as 50%) may be appropriate for larger jobs.
<property>
<name>mapred.reduce.slowstart.completed.maps</name>
<value>0</value>
</property>

Tips and Best Practices for Jobs


This section describes changes you can make at the job level.
Use the Distributed Cache to Transfer the Job JAR
Use the distributed cache to transfer the job JAR rather than using the JobConf(Class) constructor and the
JobConf.setJar() and JobConf.setJarByClass() methods.
To add JARs to the classpath, use -libjars jar1,jar2, which will copy the local JAR files to HDFS and then
use the distributed cache mechanism to make sure they are available on the task nodes and are added to the
task classpath.
The advantage of this over JobConf.setJar is that if the JAR is on a task node it won't need to be copied again
if a second task from the same job runs on that node, though it will still need to be copied from the launch
machine to HDFS.
Note: -libjars works only if your MapReduce driver uses ToolRunner. If it doesn't, you would need
to use the DistributedCache APIs (Cloudera does not recommend this).
For more information, see item 1 in the blog post How to Include Third-Party Libraries in Your MapReduce Job.
Changing the Logging Level on a Job (MRv1)
You can change the logging level for an individual job. You do this by setting the following properties in the job
configuration (JobConf):

Cloudera Administration | 279

Performance Management
mapreduce.map.log.level
mapreduce.reduce.log.level
Valid values are NONE, INFO, WARN, DEBUG, TRACE, and ALL.
Example:
JobConf conf = new JobConf();
...
conf.set("mapreduce.map.log.level", "DEBUG");
conf.set("mapreduce.reduce.log.level", "TRACE");
...

Choosing a Data Compression Format


Whether to compress your data and which compression formats to use can have a significant impact on
performance. Two of the most important places to consider data compression are in terms of MapReduce jobs
and data stored in HBase. For the most part, the principles are similar for each.
General Guidelines
You need to balance the processing capacity required to compress and uncompress the data, the disk IO
required to read and write the data, and the network bandwidth required to send the data across the network.
The correct balance of these factors depends upon the characteristics of your cluster and your data, as well
as your usage patterns.
Compression is not recommended if your data is already compressed (such as images in JPEG format). In
fact, the resulting file can actually be larger than the original.
GZIP compression uses more CPU resources than Snappy or LZO, but provides a higher compression ratio.
GZip is often a good choice for cold data, which is accessed infrequently. Snappy or LZO are a better choice
for hot data, which is accessed frequently.
BZip2 can also produce more compression than GZip for some types of files, at the cost of some speed when
compressing and decompressing. HBase does not support BZip2 compression.
Snappy often performs better than LZO. It is worth running tests to see if you detect a significant difference.
For MapReduce, if you need your compressed data to be splittable, BZip2, LZO, and Snappy formats are
splittable, but GZip is not. Splittability is not relevant to HBase data.
For MapReduce, you can compress either the intermediate data, the output, or both. Adjust the parameters
you provide for the MapReduce job accordingly. The following examples compress both the intermediate
data and the output. MR2 is shown first, followed by MR1.
MRv2
hadoop jar hadoop-examples-.jar sort "-Dmapreduce.compress.map.output=true"
"-Dmapreduce.map.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec"
"-Dmapreduce.output.compress=true"
"-Dmapreduce.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec"
-outKey
org.apache.hadoop.io.Text -outValue org.apache.hadoop.io.Text input output

MRv1
hadoop jar hadoop-examples-.jar sort "-Dmapred.compress.map.output=true"
"-Dmapred.map.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec"
"-Dmapred.output.compress=true"
"-Dmapred.output.compression.codec=org.apache.hadoop.io.compress.GzipCodec"
-outKey
org.apache.hadoop.io.Text -outValue org.apache.hadoop.io.Text input output

280 | Cloudera Administration

Performance Management
Configuring Data Compression Using Cloudera Manager
To configure support for LZO using Cloudera Manager, you must install the GPL Extras parcel, then configure
services to use it. See Installing the GPL Extras Parcel and Configuring Services to Use the GPL Extras Parcel on
page 244.
Configuring Data Compression Using the Command Line
Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.
To configure support for LZO in CDH, see Step 5: (Optional) Install LZO and Configuring LZO. Snappy support is
included in CDH.
To use Snappy in a MapReduce job, see Using Snappy for MapReduce Compression. Use the same method for
LZO, with the codec com.hadoop.compression.lzo.LzopCodec instead.
Further Reading
For more information about compression algorithms in Hadoop, see Hadoop: The Definitive Guide by Tom White.

Tuning the Solr Server


Solr performance tuning is a complex task. The following sections provide more details.

Tuning to Complete During Setup


Some tuning is best completed during the setup of you system or may require some re-indexing.
Configuring Lucene Version Requirements
You can configure Solr to use a specific version of Lucene. This can help ensure that the Lucene version that
Search uses includes the latest features and bug fixes. At the time that a version of Solr ships, Solr is typically
configured to use the appropriate Lucene version, in which case there is no need to change this setting. If a
subsequent Lucene update occurs, you can configure the Lucene version requirements by directly editing the
luceneMatchVersion element in the solrconfig.xml file. Versions are typically of the form x.y, such as 4.4.
For example, to specify version 4.4, you would ensure the following setting exists in solrconfig.xml:
<luceneMatchVersion>4.4</luceneMatchVersion>

Designing the Schema


When constructing a schema, use data types that most accurately describe the data that the fields will contain.
For example:
Use the tdate type for dates. Do this instead of representing dates as strings.
Consider using the text type that applies to your language, instead of using String. For example, you might
use text_en. Text types support returning results for subsets of an entry. For example, querying on "john"
would find "John Smith", whereas with the string type, only exact matches are returned.
For IDs, use the string type.

Cloudera Administration | 281

Performance Management
General Tuning
The following tuning categories can be completed at any time. It is less important to implement these changes
before beginning to use your system.
General Tips
Enabling multi-threaded faceting can provide better performance for field faceting. When multi-threaded
faceting is enabled, field faceting tasks are completed in a parallel with a thread working on every field
faceting task simultaneously. Performance improvements do not occur in all cases, but improvements are
likely when all of the following are true:
The system uses highly concurrent hardware.
Faceting operations apply to large data sets over multiple fields.
There is not an unusually high number of queries occurring simultaneously on the system. Systems that
are lightly loaded or that are mainly engaged with ingestion and indexing may be helped by multi-threaded
faceting; for example, a system ingesting articles and being queried by a researcher. Systems heavily
loaded by user queries are less likely to be helped by multi-threaded faceting; for example, an e-commerce
site with heavy user-traffic.
Note: Multi-threaded faceting only applies to field faceting and not to query faceting.
Field faceting identifies the number of unique entries for a field. For example, multi-threaded
faceting could be used to simultaneously facet for the number of unique entries for the fields,
"color" and "size". In such a case, there would be two threads, and each thread would work on
faceting one of the two fields.
Query faceting identifies the number of unique entries that match a query for a field. For
example, query faceting could be used to find the number of unique entries in the "size" field
are between 1 and 5. Multi-threaded faceting does not apply to these operations.
To enable multi-threaded faceting, add facet-threads to queries. For example, to use up to 1000 threads,
you might use a query as follows:
http://localhost:8983/solr/collection1/select?q=*:*&facet=true&fl=id&facet.field=f0_ws&facet.threads=1000

If facet-threads is omitted or set to 0, faceting is single-threaded. If facet-threads is set to a negative


value, such as -1, multi-threaded faceting will use as many threads as there are fields to facet up to the
maximum number of threads possible on the system.
If your environment does not require Near Real Time (NRT), turn off soft auto-commit in solrconfig.xml.
In most cases, do not change the default batch size setting of 1000. If you are working with especially large
documents, you may consider decreasing the batch size.
To help identify any garbage collector (GC) issues, enable GC logging in production. The overhead is low and
the JVM supports GC log rolling as of 1.6.0_34.
The minimum recommended GC logging flags are: -XX:+PrintGCTimeStamps -XX:+PrintGCDateStamps
-XX:+PrintGCDetails.
To rotate the GC logs: -Xloggc: -XX:+UseGCLogFileRotation -XX:NumberOfGCLogFiles=
-XX:GCLogFileSize=.
Solr and HDFS - the Block Cache
Cloudera Search enables Solr to store indexes in an HDFS filesystem. To maintain performance, an HDFS block
cache has been implemented using Least Recently Used (LRU) semantics. This enables Solr to cache HDFS index
files on read and write, storing the portions of the file in JVM "direct memory" (meaning off heap) by default or
optionally in the JVM heap. Direct memory is preferred as it is not affected by garbage collection.

282 | Cloudera Administration

Performance Management
Batch jobs typically do not make use of the cache, while Solr servers (when serving queries or indexing documents)
should. When running indexing using MapReduce, the MR jobs themselves do not make use of the block cache.
Block caching is turned off by default and should be left disabled.
Tuning of this cache is complex and best practices are continually being refined. In general, allocate a cache that
is about 10-20% of the amount of memory available on the system. For example, when running HDFS and Solr
on a host with 50 GB of memory, typically allocate 5-10 GB of memory using
solr.hdfs.blockcache.slab.count. As index sizes grow you may need to tune this parameter to maintain
optimal performance.
Note: Block cache metrics are currently unavailable.
Configuration
The following parameters control caching. They can be configured at the Solr process level by setting the respective
system property or by editing the solrconfig.xml directly.
Parameter

Default

Description

solr.hdfs.blockcache.enabled

true

Enable the block cache.

solr.hdfs.blockcache.read.enabled

true

Enable the read cache.

solr.hdfs.blockcache.write.enabled

false

Enable the write cache.

solr.hdfs.blockcache.direct.memory.allocation true

Enable direct memory allocation. If this


is false, heap is used.

solr.hdfs.blockcache.slab.count

Number of memory slabs to allocate.


Each slab is 128 MB in size.

solr.hdfs.blockcache.global

true

If enabled, a single HDFS block cache is


used for all SolrCores on a host. If
blockcache.global is disabled, each
SolrCore on a host creates its own private
HDFS block cache. Enabling this
parameter simplifies managing HDFS
block cache memory.

Note:
Increasing the direct memory cache size may make it necessary to increase the maximum direct
memory size allowed by the JVM. Each Solr slab allocates the slab's memory, which is 128 MB by
default, as well as allocating some additional direct memory overhead. Therefore, ensure that the
MaxDirectMemorySize is set comfortably above the value expected for slabs alone. The amount of
additional memory required varies according to multiple factors, but for most cases, setting
MaxDirectMemorySize to at least 20-30% more than the total memory configured for slabs is
sufficient. Setting the MaxDirectMemorySize to the number of slabs multiplied by the slab size does
not provide enough memory.
To set MaxDirectMemorySize using Cloudera Manager
1.
2.
3.
4.
5.

Go to the Solr service.


Click the Configuration tab.
In the Search box, type Java Direct Memory Size of Solr Server in Bytes.
Set the new direct memory value.
Restart Solr servers after editing the parameter.

Solr HDFS optimizes caching when performing NRT indexing using Lucene's NRTCachingDirectory.

Cloudera Administration | 283

Performance Management
Lucene caches a newly created segment if both of the following conditions are true:
The segment is the result of a flush or a merge and the estimated size of the merged segment is <=
solr.hdfs.nrtcachingdirectory.maxmergesizemb.
The total cached bytes is <= solr.hdfs.nrtcachingdirectory.maxcachedmb.
The following parameters control NRT caching behavior:
Parameter

Default

Description

solr.hdfs.nrtcachingdirectory.enable

true

Whether to enable the


NRTCachingDirectory.

solr.hdfs.nrtcachingdirectory.maxcachedmb

192

Size of the cache in megabytes.

solr.hdfs.nrtcachingdirectory.maxmergesizemb 16

Maximum segment size to cache.

Here is an example of solrconfig.xml with defaults:


<directoryFactory name="DirectoryFactory">
<bool name="solr.hdfs.blockcache.enabled">${solr.hdfs.blockcache.enabled:true}</bool>
<int
name="solr.hdfs.blockcache.slab.count">${solr.hdfs.blockcache.slab.count:1}</int>
<bool
name="solr.hdfs.blockcache.direct.memory.allocation">${solr.hdfs.blockcache.direct.memory.allocation:true}</bool>
<int
name="solr.hdfs.blockcache.blocksperbank">${solr.hdfs.blockcache.blocksperbank:16384}</int>
<bool
name="solr.hdfs.blockcache.read.enabled">${solr.hdfs.blockcache.read.enabled:true}</bool>
<bool
name="solr.hdfs.blockcache.write.enabled">${solr.hdfs.blockcache.write.enabled:true}</bool>
<bool
name="solr.hdfs.nrtcachingdirectory.enable">${solr.hdfs.nrtcachingdirectory.enable:true}</bool>
<int
name="solr.hdfs.nrtcachingdirectory.maxmergesizemb">${solr.hdfs.nrtcachingdirectory.maxmergesizemb:16}</int>
<int
name="solr.hdfs.nrtcachingdirectory.maxcachedmb">${solr.hdfs.nrtcachingdirectory.maxcachedmb:192}</int>
</directoryFactory>

The following example illustrates passing Java options by editing the /etc/default/solr or
/opt/cloudera/parcels/CDH-*/etc/default/solr configuration file:
CATALINA_OPTS="-Xmx10g -XX:MaxDirectMemorySize=20g -XX:+UseLargePages
-Dsolr.hdfs.blockcache.slab.count=100"

For better performance, Cloudera recommends setting the Linux swap space on all Solr server hosts as shown
below:
# minimize swappiness
sudo sysctl vm.swappiness=10
sudo bash -c 'echo "vm.swappiness=10">> /etc/sysctl.conf'
# disable swap space until next reboot:
sudo /sbin/swapoff -a

Cloudera previously recommended a setting of 0, but in recent kernels (such as those included with RedHat 6.4
and higher, and Ubuntu 12.04 LTS and higher) a setting of 0 might lead to out of memory issues per this blog
post: http://www.percona.com/blog/2014/04/28/oom-relation-vm-swappiness0-new-kernel/.

284 | Cloudera Administration

Performance Management
Threads
Configure the Tomcat server to have more threads per Solr instance. Note that this is only effective if your
hardware is sufficiently powerful to accommodate the increased threads. 10,000 threads is a reasonable number
to try in many cases.
To change the maximum number of threads, add a maxThreads element to the Connector definition in the
Tomcat server's server.xml configuration file. For example, if you installed Search for CDH 5 using parcels
installation, you would modify the Connector definition in the <parcel
path>/CDH/etc/solr/tomcat-conf.dist/conf/server.xml file so this:
<Connector port="${solr.port}" protocol="HTTP/1.1"
connectionTimeout="20000"
redirectPort="8443" />

Becomes this:
<Connector port="${solr.port}" protocol="HTTP/1.1"
maxThreads="10000"
connectionTimeout="20000"
redirectPort="8443" />

Garbage Collection
Choose different garbage collection options for best performance in different environments. Some garbage
collection options typically chosen include:
Concurrent low pause collector: Use this collector in most cases. This collector attempts to minimize "Stop
the World" events. Avoiding these events can reduce connection timeouts, such as with ZooKeeper, and may
improve user experience. This collector is enabled using -XX:+UseConcMarkSweepGC.
Throughput collector: Consider this collector if raw throughput is more important than user experience. This
collector typically uses more "Stop the World" events so this may negatively affect user experience and
connection timeouts such as ZooKeeper heartbeats. This collector is enabled using -XX:+UseParallelGC.
If UseParallelGC "Stop the World" events create problems, such as ZooKeeper timeouts, consider using the
UseParNewGC collector as an alternative collector with similar throughput benefits.
You can also affect garbage collection behavior by increasing the Eden space to accommodate new objects. With
additional Eden space, garbage collection does not need to run as frequently on new objects.
Replicated Data
You can adjust the degree to which different data is replicated.
Replicas
If you have sufficient additional hardware, add more replicas for a linear boost of query throughput. Note that
adding replicas may slow write performance on the first replica, but otherwise this should have minimal negative
consequences.
Transaction Log Replication
Beginning with CDH 5.4.1, Search for CDH supports configurable transaction log replication levels for replication
logs stored in HDFS.
Configure the replication factor by modifying the tlogDfsReplication setting in solrconfig.xml. The
tlogDfsReplication is a new setting in the updateLog settings area. An excerpt of the solrconfig.xml file
where the transaction log replication factor is set is as follows:
<updateHandler class="solr.DirectUpdateHandler2">
<!-- Enables a transaction log, used for real-time get, durability, and
and solr cloud replica recovery. The log can grow as big as
uncommitted changes to the index, so use of a hard autoCommit

Cloudera Administration | 285

Performance Management
is recommended (see below).
"dir" - the target directory for transaction logs, defaults to the
solr data directory. -->
<updateLog>
<str name="dir">${solr.ulog.dir:}</str>
<int name="tlogDfsReplication">3</int>
</updateLog>

You might want to increase the replication level from the default level of 1 to some higher value such as 3.
Increasing the transaction log replication level can:
Reduce the chance of data loss, especially when the system is otherwise configured to have single replicas
of shards. For example, having single replicas of shards is reasonable when autoAddReplicas is enabled,
but without additional transaction log replicas, the risk of data loss during a node failure would increase.
Facilitate rolling upgrade of HDFS while Search is running. If you have multiple copies of the log, when a node
with the transaction log becomes unavailable during the rolling upgrade process, another copy of the log can
continue to collect transactions.
Facilitate HDFS write lease recovery.
Initial testing shows no significant performance regression for common use cases.
Shards
In some cases, oversharding can help improve performance including intake speed. If your environment includes
massively parallel hardware and you want to use these available resources, consider oversharding. You might
increase the number of replicas per host from 1 to 2 or 3. Making such changes creates complex interactions,
so you should continue to monitor your system's performance to ensure that the benefits of oversharding do
not outweigh the costs.
Commits
Changing commit values may improve performance in some situation. These changes result in tradeoffs and
may not be beneficial in all cases.
For hard commit values, the default value of 60000 (60 seconds) is typically effective, though changing this
value to 120 seconds may improve performance in some cases. Note that setting this value to higher values,
such as 600 seconds may result in undesirable performance tradeoffs.
Consider increasing the auto-commit value from 15000 (15 seconds) to 120000 (120 seconds).
Enable soft commits and set the value to the largest value that meets your requirements. The default value
of 1000 (1 second) is too aggressive for some environments.

Other Resources
General information on Solr caching is available on the SolrCaching page on the Solr Wiki.
Information on issues that influence performance is available on the SolrPerformanceFactors page on the
Solr Wiki.
Resource Management describes how to use Cloudera Manager to manage resources, for example with Linux
cgroups.
For information on improving querying performance, see How to make searching faster.
For information on improving indexing performance, see How to make indexing faster.

Tuning Spark Applications


Optimizing Spark Performance
CDH 5.3 introduces a performance optimization that causes Spark to prefer RDDs that are already cached locally
in HDFS. Using locally cached RDDs is important enough that Spark will wait a short time for the executors near
286 | Cloudera Administration

Performance Management
these caches to be free. Spark does not start executors on hosts with cached data, and there is no further chance
to select them during the task-matching phase. This is not a problem for most workloads, since most workloads
start executors on most or all hosts in the cluster. However, if you do have problems with the optimization, the
following constructor:
@DeveloperApi
def this(config: SparkConf, preferredNodeLocationData: Map[ String, Set[SplitInfo]])
= {
this(config)
this.preferredNodeLocationData = preferredNodeLocationData
}

allows you to explicitly specify the preferred locations to start executors. The following example from
examples/src/main/scala/org/apache/spark/examples/SparkHdfsLR.scala, provides an example of
using this API:
...
val
val
val
val

sparkConf = new SparkConf().setAppName("SparkHdfsLR")


inputPath = args(0)
conf = SparkHadoopUtil.get.newConfiguration()
sc = new SparkContext(sparkConf,
InputFormatInfo.computePreferredLocations(
Seq(new InputFormatInfo(conf, classOf[org.apache.hadoop.mapred.TextInputFormat],
inputPath))
...

Cloudera Administration | 287

High Availability

High Availability
This guide is for Apache Hadoop system administrators who want to enable continuous availability by configuring
clusters without single points of failure.

HDFS High Availability


This section provides an overview of the HDFS high availability (HA) feature and how to configure and manage
an HA HDFS cluster.

Introduction to HDFS High Availability


This section assumes that the reader has a general understanding of components in an HDFS cluster. For details,
see the Apache HDFS Architecture Guide.

Background
In a standard configuration, the NameNode is a single point of failure (SPOF) in an HDFS cluster. Each cluster
has a single NameNode, and if that host or process became unavailable, the cluster as a whole is unavailable
until the NameNode is either restarted or brought up on a new host. The Secondary NameNode does not provide
failover capability.
The standard configuration reduces the total availability of an HDFS cluster in two major ways:
In the case of an unplanned event such as a host crash, the cluster is unavailable until an operator restarts
the NameNode.
Planned maintenance events such as software or hardware upgrades on the NameNode machine result in
periods of cluster downtime.
HDFS HA addresses the above problems by providing the option of running two NameNodes in the same cluster,
in an active/passive configuration. These are referred to as the active NameNode and the standby NameNode.
Unlike the Secondary NameNode, the standby NameNode is hot standby, allowing a fast automatic failover to
a new NameNode in the case that a host crashes, or a graceful administrator-initiated failover for the purpose
of planned maintenance. You cannot have more than two NameNodes.

Implementation
Cloudera Manager 5 and CDH 5 support Quorum-based Storage on page 289 as the only HA implementation. In
contrast, CDH 4 supports Quorum-based Storage and shared storage using NFS. For instructions on switching
to Quorum-based storage, see Converting From an NFS-mounted Shared Edits Directory to Quorum-based
Storage on page 314.

288 | Cloudera Administration

High Availability
Important:
If you have upgraded from Cloudera Manager 4, and have a CDH 4 cluster using HA with an
NFS-mounted shared edits directory, your HA configuration will continue to work. However, you
will see a validation warning recommending you switch to Quorum-based storage.
If you are using NFS-mounted shared edits directories and you disable HA, you will not be able to
re-enable HA in Cloudera Manager 5 using NFS-mounted shared directories. Instead, you should
configure HA to use Quorum-based storage.
If you have HA enabled using an NFS-mounted shared edits directory, you will be blocked from
upgrading CDH 4 to CDH 5. You must disable HA in order to perform the upgrade. After the upgrade,
you will not be able to use NFS-mounted shared edits directories for edit log storage.
If you are using CDH 4: CDH 4 did not support Quorum-based Storage. Therefore, if you were using
a HA configuration and you disable it, you will not be able to enable it through Cloudera Manager
5, since Cloudera Manager 5 does not support NFS-mounted storage. It is recommended that you
upgrade your CDH 4 deployment to a more recent version of CDH.

Quorum-based Storage

Quorum-based Storage refers to the HA implementation that uses a Quorum Journal Manager (QJM).
In order for the standby NameNode to keep its state synchronized with the active NameNode in this
implementation, both nodes communicate with a group of separate daemons called JournalNodes. When any
namespace modification is performed by the active NameNode, it durably logs a record of the modification to a
majority of the JournalNodes. The standby NameNode is capable of reading the edits from the JournalNodes,
and is constantly watching them for changes to the edit log. As the standby Node sees the edits, it applies them
to its own namespace. In the event of a failover, the standby ensures that it has read all of the edits from the
JournalNodes before promoting itself to the active state. This ensures that the namespace state is fully
synchronized before a failover occurs.
In order to provide a fast failover, it is also necessary that the standby NameNode has up-to-date information
regarding the location of blocks in the cluster. In order to achieve this, DataNodes are configured with the location
of both NameNodes, and they send block location information and heartbeats to both.
It is vital for the correct operation of an HA cluster that only one of the NameNodes be active at a time. Otherwise,
the namespace state would quickly diverge between the two, risking data loss or other incorrect results. In order
to ensure this property and prevent the so-called "split-brain scenario," JournalNodes only ever allow a single
NameNode to be a writer at a time. During a failover, the NameNode which is to become active simply takes
over the role of writing to the JournalNodes, which effectively prevents the other NameNode from continuing
in the active state, allowing the new active NameNode to safely proceed with failover.
Note: Because of this, fencing is not required, but it is still useful; see Enabling HDFS HA on page 291.
Shared Storage Using NFS
In order for the standby NameNode to keep its state synchronized with the active NameNode, this implementation
requires that the two nodes both have access to a directory on a shared storage device (for example, an NFS
mount from a NAS).
When any namespace modification is performed by the active NameNode, it durably logs a record of the
modification to an edit log file stored in the shared directory. The standby NameNode constantly watches this
directory for edits, and when edits occur, the standby NameNode applies them to its own namespace. In the
event of a failover, the standby will ensure that it has read all of the edits from the shared storage before
promoting itself to the active state. This ensures that the namespace state is fully synchronized before a failover
occurs.
In order to prevent the "split-brain scenario" in which the namespace state is diverged between the two
NameNodes, an administrator must configure at least one fencing method for the shared storage. If, during a
Cloudera Administration | 289

High Availability
failover, it cannot be verified that the previous active NameNode has relinquished its active state, the fencing
process is responsible for cutting off the previous active NameNode's access to the shared edits storage. This
prevents it from making any further edits to the namespace, allowing the new active NameNode to safely proceed
with failover.
Automatic Failover Implementation
Automatic failover relies on two additional components in an HDFS: a ZooKeeper quorum, and the
ZKFailoverController process (abbreviated as ZKFC). In Cloudera Manager, the ZKFC process maps to the
HDFS Failover Controller role.
Apache ZooKeeper is a highly available service for maintaining small amounts of coordination data, notifying
clients of changes in that data, and monitoring clients for failures. The implementation of HDFS automatic
failover relies on ZooKeeper for the following functions:
Failure detection - each of the NameNode machines in the cluster maintains a persistent session in ZooKeeper.
If the machine crashes, the ZooKeeper session will expire, notifying the other NameNode that a failover
should be triggered.
Active NameNode election - ZooKeeper provides a simple mechanism to exclusively elect a node as active.
If the current active NameNode crashes, another node can take a special exclusive lock in ZooKeeper indicating
that it should become the next active NameNode.
The ZKFailoverController (ZKFC) is a ZooKeeper client that also monitors and manages the state of the
NameNode. Each of the hosts that run a NameNode also run a ZKFC. The ZKFC is responsible for:
Health monitoring - the ZKFC contacts its local NameNode on a periodic basis with a health-check command.
So long as the NameNode responds promptly with a healthy status, the ZKFC considers the NameNode
healthy. If the NameNode has crashed, frozen, or otherwise entered an unhealthy state, the health monitor
marks it as unhealthy.
ZooKeeper session management - when the local NameNode is healthy, the ZKFC holds a session open in
ZooKeeper. If the local NameNode is active, it also holds a special lock znode. This lock uses ZooKeeper's
support for "ephemeral" nodes; if the session expires, the lock node is automatically deleted.
ZooKeeper-based election - if the local NameNode is healthy, and the ZKFC sees that no other NameNode
currently holds the lock znode, it will itself try to acquire the lock. If it succeeds, then it has "won the election",
and is responsible for running a failover to make its local NameNode active. The failover process is similar
to the manual failover described above: first, the previous active is fenced if necessary, and then the local
NameNode transitions to active state.

Configuring Hardware for HDFS HA


In order to deploy an HA cluster using Quorum-based Storage, you should prepare the following:
NameNode hosts - These are the hosts on which you run the active and standby NameNodes. They should
have equivalent hardware to each other, and equivalent hardware to what would be used in a non-HA cluster.
JournalNode hosts - These are the hosts on which you run the JournalNodes. Cloudera recommends that
you deploy the JournalNode daemons on the "master" host or hosts (NameNode, Standby NameNode,
JobTracker, and so on) so the JournalNodes' local directories can use the reliable local storage on those
machines.
If co-located on the same host, each JournalNode process and each NameNode process should have its own
dedicated disk. You should not use SAN or NAS storage for these directories.
There must be at least three JournalNode daemons, since edit log modifications must be written to a majority
of JournalNodes. This will allow the system to tolerate the failure of a single host. You can also run more
than three JournalNodes, but in order to actually increase the number of failures the system can tolerate,
you should run an odd number of JournalNodes, (three, five, seven, and so on). Note that when running with
N JournalNodes, the system can tolerate at most (N - 1) / 2 failures and continue to function normally. If the

290 | Cloudera Administration

High Availability
requisite quorum is not available, the NameNode will not format or start, and you will see an error similar to
this:
12/10/01 17:34:18 WARN namenode.FSEditLog: Unable to determine input streams from QJM
to [10.0.1.10:8485, 10.0.1.10:8486, 10.0.1.10:8487]. Skipping.
java.io.IOException: Timed out waiting 20000ms for a quorum of nodes to respond.

Note: In an HA cluster, the standby NameNode also performs checkpoints of the namespace state,
and thus it is not necessary to run a Secondary NameNode, CheckpointNode, or BackupNode in an
HA cluster. In fact, to do so would be an error. If you are reconfiguring a non-HA-enabled HDFS cluster
to be HA-enabled, you can reuse the hardware which you had previously dedicated to the Secondary
NameNode.

Enabling HDFS HA
An HDFS high availability (HA) cluster uses two NameNodesan active NameNode and a standby NameNode.
Only one NameNode can be active at any point in time. HDFS HA depends on maintaining a log of all namespace
modifications in a location available to both NameNodes, so that in the event of a failure, the standby NameNode
has up-to-date information about the edits and location of blocks in the cluster. For CDH 4 HA features, see the
CDH 4 High Availability Guide.
Important: Enabling and disabling HA causes a service outage for the HDFS service and all services
that depend on HDFS. Before enabling or disabling HA, ensure that there are no jobs running on your
cluster.

Enabling HDFS HA Using Cloudera Manager


Required Role:
You can use Cloudera Manager to configure your CDH 4 or CDH 5 cluster for HDFS HA and automatic failover. In
Cloudera Manager 5, HA is implemented using Quorum-based storage. Quorum-based storage relies upon a
set of JournalNodes, each of which maintains a local edits directory that logs the modifications to the namespace
metadata. Enabling HA enables automatic failover as part of the same command.
Important:
Enabling or disabling HA causes the previous monitoring history to become unavailable.
Some parameters will be automatically set as follows once you have enabled JobTracker HA. If
you want to change the value from the default for these parameters, use an advanced configuration
snippet.

mapred.jobtracker.restart.recover: true
mapred.job.tracker.persist.jobstatus.active: true
mapred.ha.automatic-failover.enabled: true
mapred.ha.fencing.methods: shell(/bin/true)

Enabling High Availability and Automatic Failover


The Enable High Availability workflow leads you through adding a second (standby) NameNode and configuring
JournalNodes. During the workflow, Cloudera Manager creates a federated namespace.
1. Perform all the configuration and setup tasks described under Configuring Hardware for HDFS HA on page
290.
2. Ensure that you have a ZooKeeper service.
3. Go to the HDFS service.

Cloudera Administration | 291

High Availability
4. Select Actions > Enable High Availability. A screen showing the hosts that are eligible to run a standby
NameNode and the JournalNodes displays.
a. Specify a name for the nameservice or accept the default name nameservice1 and click Continue.
b. In the NameNode Hosts field, click Select a host. The host selection dialog displays.
c. Check the checkbox next to the hosts where you want the standby NameNode to be set up and click OK.
The standby NameNode cannot be on the same host as the active NameNode, and the host that is chosen
should have the same hardware configuration (RAM, disk space, number of cores, and so on) as the active
NameNode.
d. In the JournalNode Hosts field, click Select hosts. The host selection dialog displays.
e. Check the checkboxes next to an odd number of hosts (a minimum of three) to act as JournalNodes and
click OK. JournalNodes should be hosted on hosts with similar hardware specification as the NameNodes.
It is recommended that you put a JournalNode each on the same hosts as the active and standby
NameNodes, and the third JournalNode on similar hardware, such as the JobTracker.
f. Click Continue.
g. In the JournalNode Edits Directory property, enter a directory location for the JournalNode edits directory
into the fields for each JournalNode host.
You may enter only one directory for each JournalNode. The paths do not need to be the same on every
JournalNode.
The directories you specify should be empty, and must have the appropriate permissions.
h. Extra Options: Decide whether Cloudera Manager should clear existing data in ZooKeeper, standby
NameNode, and JournalNodes. If the directories are not empty (for example, you are re-enabling a previous
HA configuration), Cloudera Manager will not automatically delete the contentsyou can select to delete
the contents by keeping the default checkbox selection. The recommended default is to clear the directories.
If you choose not to do so, the data should be in sync across the edits directories of the JournalNodes
and should have the same version data as the NameNodes.
i. Click Continue.
Cloudera Manager executes a set of commands that will stop the dependent services, delete, create, and
configure roles and directories as appropriate, create a nameservice and failover controller, and restart the
dependent services and deploy the new client configuration.
5. If you want to use other services in a cluster with HA configured, follow the procedures in Configuring Other
CDH Components to Use HDFS HA on page 308.
6. If you are running CDH 4.0 or 4.1, the standby NameNode may fail at the bootstrapStandby command with
the error Unable to read transaction ids 1-7 from the configured shared edits storage. Use
rsync or a similar tool to copy the contents of the dfs.name.dir directory from the active NameNode to
the standby NameNode and start the standby NameNode.

292 | Cloudera Administration

High Availability
Important: If you change the NameNode Service RPC Port (dfs.namenode.servicerpc-address)
while automatic failover is enabled, this will cause a mismatch between the NameNode address saved
in the ZooKeeper /hadoop-ha znode and the NameNode address that the Failover Controller is
configured with. This will prevent the Failover Controllers from restarting. If you need to change the
NameNode Service RPC Port after Auto Failover has been enabled, you must do the following to
re-initialize the znode:
1. Stop the HDFS service.
2. Configure the service RPC port:
a.
b.
c.
d.
e.

Go to the HDFS service.


Click the Configuration tab.
Select Scope > NameNode.
Select Category > Ports and Addresses.
Locate the NameNode Service RPC Port property or search for it by typing its name in the
Search box.
f. Change the port value as needed.
If more than one role group applies to this configuration, edit the value for the appropriate role
group. See Modifying Configuration Properties Using Cloudera Manager on page 10.
3. On a ZooKeeper server host, run zookeeper-client:
4. Execute the following to remove the configured nameservice. This example assumes the name
of the nameservice is nameservice1. You can identify the nameservice from the Federation and
High Availability section on the HDFS Instances tab:
rmr /hadoop-ha/nameservice1

5. Click the Instances tab.


6. Select Actions > Initialize High Availability State in ZooKeeper.
7. Start the HDFS service.
Fencing Methods
To ensure that only one NameNode is active at a time, a fencing method is required for the shared edits directory.
During a failover, the fencing method is responsible for ensuring that the previous active NameNode no longer
has access to the shared edits directory, so that the new active NameNode can safely proceed writing to it.
By default, Cloudera Manager configures HDFS to use a shell fencing method
(shell(./cloudera_manager_agent_fencer.py)) that takes advantage of the Cloudera Manager Agent.
However, you can configure HDFS to use the sshfence method, or you can add your own shell fencing scripts,
instead of or in addition to the one Cloudera Manager provides.
The fencing parameters are found in the Service-Wide > High Availability category under the configuration
properties for your HDFS service.
For details of the fencing methods supplied with CDH 5, and how fencing is configured, see Fencing Configuration
on page 297.

Enabling HDFS HA Using the Command Line


Important:
If you use Cloudera Manager, do not use these command-line instructions.
This information applies specifically to CDH 5.5.x. If you use an earlier version of CDH, see the
documentation for that version located at Cloudera Documentation.

Cloudera Administration | 293

High Availability
This section describes the software configuration required for HDFS HA in CDH 5 and explains how to set
configuration properties and use the command line to deploy HDFS HA.
Configuring Software for HDFS HA
Configuration Overview
As with HDFS federation configuration, HA configuration is backward compatible and allows existing single
NameNode configurations to work without change. The new configuration is designed such that all the nodes
in the cluster can have the same configuration without the need for deploying different configuration files to
different machines based on the type of the node.
HA clusters reuse the Nameservice ID to identify a single HDFS instance that may consist of multiple HA
NameNodes. In addition, there is a new abstraction called NameNode ID. Each distinct NameNode in the cluster
has a different NameNode ID. To support a single configuration file for all of the NameNodes, the relevant
configuration parameters include the Nameservice ID as well as the NameNode ID.
Changes to Existing Configuration Parameters
The following configuration parameter has changed for YARN implementations:
fs.defaultFS - formerly fs.default.name, the default path prefix used by the Hadoop FS client when none
is given. (fs.default.name is deprecated for YARN implementations, but will still work.)

Optionally, you can configure the default path for Hadoop clients to use the HA-enabled logical URI. For example,
if you use mycluster as the Nameservice ID as shown below, this will be the value of the authority portion of
all of your HDFS paths. You can configure the default path in your core-site.xml file:
For YARN:
<property>
<name>fs.defaultFS</name>
<value>hdfs://mycluster</value>
</property>

For MRv1:
<property>
<name>fs.default.name</name>
<value>hdfs://mycluster</value>
</property>

New Configuration Parameters


To configure HA NameNodes, you must add several configuration options to your hdfs-site.xml configuration
file.
The order in which you set these configurations is unimportant, but the values you choose for dfs.nameservices
and dfs.ha.namenodes.[Nameservice ID] will determine the keys of those that follow. This means that you
should decide on these values before setting the rest of the configuration options.
Configure dfs.nameservices
dfs.nameservices - the logical name for this new nameservice

Choose a logical name for this nameservice, for example mycluster, and use this logical name for the value of
this configuration option. The name you choose is arbitrary. It will be used both for configuration and as the
authority component of absolute HDFS paths in the cluster.

294 | Cloudera Administration

High Availability
Note: If you are also using HDFS federation, this configuration setting should also include the list of
other Nameservices, HA or otherwise, as a comma-separated list.
<property>
<name>dfs.nameservices</name>
<value>mycluster</value>
</property>

Configure dfs.ha.namenodes.[nameservice ID]


dfs.ha.namenodes.[nameservice ID] - unique identifiers for each NameNode in the nameservice

Configure a list of comma-separated NameNode IDs. This will be used by DataNodes to determine all the
NameNodes in the cluster. For example, if you used mycluster as the NameService ID previously, and you
wanted to use nn1 and nn2 as the individual IDs of the NameNodes, you would configure this as follows:
<property>
<name>dfs.ha.namenodes.mycluster</name>
<value>nn1,nn2</value>
</property>

Note: In this release, you can configure a maximum of two NameNodes per nameservice.
Configure dfs.namenode.rpc-address.[nameservice ID]
dfs.namenode.rpc-address.[nameservice ID].[name node ID] - the fully-qualified RPC address for each

NameNode to listen on
For both of the previously-configured NameNode IDs, set the full address and RPC port of the NameNode process.
Note that this results in two separate configuration options. For example:
<property>
<name>dfs.namenode.rpc-address.mycluster.nn1</name>
<value>machine1.example.com:8020</value>
</property>
<property>
<name>dfs.namenode.rpc-address.mycluster.nn2</name>
<value>machine2.example.com:8020</value>
</property>

Note: If necessary, you can similarly configure the servicerpc-address setting.


Configure dfs.namenode.http-address.[nameservice ID]
dfs.namenode.http-address.[nameservice ID].[name node ID] - the fully-qualified HTTP address for

each NameNode to listen on


Similarly torpc-addressabove, set the addresses for both NameNodes' HTTP servers to listen on. For example:
<property>
<name>dfs.namenode.http-address.mycluster.nn1</name>
<value>machine1.example.com:50070</value>
</property>
<property>
<name>dfs.namenode.http-address.mycluster.nn2</name>
<value>machine2.example.com:50070</value>
</property>

Note: If you have Hadoop Kerberos security features enabled, and you intend to use HSFTP, you
should also set the https-address similarly for each NameNode.
Configure dfs.namenode.shared.edits.dir
Cloudera Administration | 295

High Availability
dfs.namenode.shared.edits.dir - the location of the shared storage directory

Configure the addresses of the JournalNodes which provide the shared edits storage, written to by the Active
NameNode and read by the Standby NameNode to stay up-to-date with all the file system changes the Active
NameNode makes. Though you must specify several JournalNode addresses, you should only configure one of
these URIs. The URI should be in the form:
qjournal://<host1:port1>;<host2:port2>;<host3:port3>/<journalId>

The Journal ID is a unique identifier for this nameservice, which allows a single set of JournalNodes to provide
storage for multiple federated namesystems. Though it is not a requirement, it's a good idea to reuse the
Nameservice ID for the journal identifier.
For example, if the JournalNodes for this cluster were running on the machines node1.example.com,
node2.example.com, and node3.example.com, and the nameservice ID were mycluster, you would use the
following as the value for this setting (the default port for the JournalNode is 8485):
<property>
<name>dfs.namenode.shared.edits.dir</name>
<value>qjournal://node1.example.com:8485;node2.example.com:8485;node3.example.com:8485/mycluster</value>
</property>

Configure dfs.journalnode.edits.dir
dfs.journalnode.edits.dir - the path where the JournalNode daemon will store its local state

On each JournalNode machine, configure the absolute path where the edits and other local state information
used by the JournalNodes will be stored; use only a single path per JournalNode. (The other JournalNodes provide
redundancy; you can also configure this directory on a locally-attached RAID-1 or RAID-10 array.)
For example:
<property>
<name>dfs.journalnode.edits.dir</name>
<value>/data/1/dfs/jn</value>
</property>

Now create the directory (if it doesn't already exist) and make sure its owner is hdfs, for example:
$ sudo mkdir -p /data/1/dfs/jn
$ sudo chown -R hdfs:hdfs /data/1/dfs/jn

Client Failover Configuration


dfs.client.failover.proxy.provider.[nameservice ID] - the Java class that HDFS clients use to contact

the Active NameNode


Configure the name of the Java class which the DFS client will use to determine which NameNode is the current
active, and therefore which NameNode is currently serving client requests. The only implementation which
currently ships with Hadoop is the ConfiguredFailoverProxyProvider, so use this unless you are using a
custom one. For example:
<property>
<name>dfs.client.failover.proxy.provider.mycluster</name>
<value>org.apache.hadoop.hdfs.server.namenode.ha.ConfiguredFailoverProxyProvider</value>
</property>

296 | Cloudera Administration

High Availability
Fencing Configuration
dfs.ha.fencing.methods - a list of scripts or Java classes which will be used to fence the active NameNode

during a failover
It is desirable for correctness of the system that only one NameNode be in the active state at any given time.
Important: When you use Quorum-based Storage, only one NameNode will ever be allowed to write
to the JournalNodes, so there is no potential for corrupting the file system metadata in a "split-brain"
scenario. But when a failover occurs, it is still possible that the previously active NameNode could
serve read requests to clients - and these requests may be out of date - until that NameNode shuts
down when it tries to write to the JournalNodes. For this reason, it is still desirable to configure some
fencing methods even when using Quorum-based Storage.
To improve the availability of the system in the event the fencing mechanisms fail, it is advisable to configure
a fencing method which is guaranteed to return success as the last fencing method in the list.
Note: If you choose to use no actual fencing methods, you still must configure something for this
setting, for example shell(/bin/true).
The fencing methods used during a failover are configured as a carriage-return-separated list, and these will
be attempted in order until one of them indicates that fencing has succeeded.
There are two fencing methods which ship with Hadoop:
sshfence
shell
For information on implementing your own custom fencing method, see the org.apache.hadoop.ha.NodeFencer
class.
Configuring the sshfence fencing method
sshfence - SSH to the active NameNode and kill the process

The sshfence option uses SSH to connect to the target node and uses fuser to kill the process listening on the
service's TCP port. In order for this fencing option to work, it must be able to SSH to the target node without
providing a passphrase. Thus, you must also configure the dfs.ha.fencing.ssh.priv